International journal of computer integrated manufacturing , tập 23, số 12, 2010

International Journal of COMPUTER INTEGRATED MANUFACTURING
EDITOR-IN-CHIEF

Stephen T. Newman
Department of Mechanical Engineering
University of Bath,
Bath BA2 7AY, UK
e-mail:
EDITOR: ASIA (PACIFIC)

EDITOR: NORTH AMERICA

George Q. Huang

Paul Kenneth Wright

Department of Industrial and
Manufacturing Systems
Engineering, The University of
Hong Kong,
Pokfulam Road, Hong Kong,
China
e-mail:

Department of Mechanical
Engineering
University of California, 5133 Etcheverry
Hall
Berkeley, CA 94720-1740, USA
e-mail:

EDITOR: NORTH AMERICA

Paul G. Ranky

MANAGING EDITOR and
BOOK REVIEWS EDITOR

Aydin Nassehi
Department of Mechanical Engineering,
University of Bath, Bath, BA2 7AY, UK
e-mail:
Founding Editor: David Hughes

Department of Industrial and
Manufacturing Engineering
New Jersey Institute of Technology
University Heights, Newark NJ 07102, USA
e-mail:

EDITOR: EUROPE

Franc¸ois Vernadat
IT & Telecommunications Division
European Court of Auditors
12, Rue Alcide de Gasperi
1615 Luxembourg
e-mail:
CONSULTING EDITOR

CONSULTING EDITOR

George Chryssolouris

David J. Williams*

Laboratory for Manufacturing
Systems, Department of
Mechanical Engineering and
Aeronautics,
University of Patras, Patras 26110, Greece
e-mail:

Wolfson School of Mechanical and
Manufacturing
Engineering, Loughborough
University,
Loughborough, Leics LE11 3TU, UK
e-mail:

*CIRP Representative

INTERNATIONAL EDITORIAL BOARD
C. R. Boe¨r
ICIMSI Istituto CIM della Svizzera Italiano,
Switzerland
J. Browne
National University of Ireland, Galway
F. T. S. Chan
The Hong Kong Polytechnic University, China

B.-K. Choi
KAIST, South Korea
I. Choi
Pohang University of Science and Technology,
South Korea
H. A. ElMaraghy
University of Windsor, Ontario, Canada
Y. Fan
Tsinghua University, China
P. Ferreira
University of Illinois at Urbana-Champaign,
USA
Y. Furukawa
Tokyo Metropolitan University, Japan
N. N. Z. Gindy
University of Nottingham, UK
A. Gunasekaran
University of Massachusetts, USA
K. Hitomi
Kyoto University, Japan

S. Hobbs
Delcam Plc, Birmingham, UK
S.-L. Hwang
National Tsing Hua University, Taiwan
M. Jeng
National Taiwan Ocean University, Taiwan
A. Jones
National Institute of Standards
& Technology, USA

S. Joshi
Pennsylvania State University, USA
F. Jovane
Politecnico di Milano, Italy
H. Katayama
Waseda University, Japan
K. Kosanke
CIMOSA, Germany
A. Kusiak
University of Iowa, USA
A. Molina
ITESM Monterrey, Mexico
L. B. Newnes
University of Bath, UK
S. Y. Nof
Purdue University, USA
P. O’Grady
University of Iowa, USA

P. Pokorny
Technical University of Liberec, Czech Republic
P. Rogers
University of Calgary, Canada
I. Sabuncuoglu
Bilkent University, Turkey
M. K. Tiwari
Indian Institute of Technology, Kharagpur,
India
R. Uzsoy
Purdue University, USA

H. Van Brussel
Katholieke Universiteit TE Leuven, Belgium
H. Van Dyke Parunak
ERIM Center for Electronic Commerce, USA
J. Vancza
MTA Sztaki, Hungary
R. Veeramani
University of Wisconsin-Madison, USA
L. Wang
University of Sko¨vde, Sweden
H.-J. Warnecke
Fraunhofer-IPA, Germany
R. H. Weston
Loughborough University, UK
X. Xu
University of Auckland, New Zealand


International Journal of Computer Integrated Manufacturing
Vol. 23, No. 12, December 2010, 1059–1070

Towards expressive ontology-based approaches to manufacturing knowledge
representation and sharing
Nitishal Chungooraa*, Osiris Canciglieri Jr.b and R.I.M. Younga
a

Wolfson School of Mechanical and Manufacturing Engineering, Loughborough University, Loughborough, UK; bLaboratory of
Automation and Systems (LAS), Pontifical Catholic University of Parana´ (PUCPR), Rua Imaculada Conceic¸a˜o, 1155 Prado
Velho, Curitiba, PR, Brazil, CEP 80215-030
(Received 25 April 2010; final version received 24 August 2010)

The present capability that ontological approaches offer to formally represent and share manufacturing knowledge
is dependent on the choice of ontological formalism. Currently, there exists a spectrum of these formalisms, which is
being subjectively exploited across multiple domains in design and manufacture. Hence, there is an important
prerequisite to achieve an understanding of which family of formalism strictly enables the expressive capture of
semantics to progress towards meaningful information and viable knowledge sharing. This article analyses the
relative strengths and weaknesses in employing a ‘lightweight’ ontology versus a ‘heavyweight’ version of the
ontology to represent and share knowledge between multiple domains in injection moulding design and
manufacture. A pertinent direction, from an ontology perspective, is then exposed as a prescription for the
improved capture and dissemination of formal semantics, to support multi-domain knowledge sharing.
Keywords: design and manufacture; lightweight ontology; heavyweight ontology; semantics; knowledge sharing;
injection moulding

1.

Introduction

Ontological approaches are nowadays increasingly
being applied to support the formal capture and
sharing of the meaning and intent (i.e. semantics) of
design and manufacture concepts. For a particular
domain, the representation of the required semantics is
held in an ontology and the knowledge base (KB)
deployed from the ontology is used to populate
knowledge which should consistently derive from the
semantic structures within the ontology. Represented
knowledge in a KB provides useful support for key
engineering decisions, for example the ways in which a
designer’s intent in the design domain could affect the
selection of manufacturing processes in the manufacturing domain. Thus, expressive manufacturing knowledge refers to populated knowledge in a KB, based on
the unambiguous definition of semantics structures,

which carry enriched formal meaning.
Unfortunately at present, the seamless exchange of
design and manufacture semantics for knowledge
sharing is still not achievable as a result of domain
models that do not carry sufficiently expressive
semantics. This is because there are currently several
ontological formalisms, of varying expressiveness (Ray
2004) and system interaction capabilities, which do not
all necessarily address the knowledge capture and

*Corresponding author. Email:
ISSN 0951-192X print/ISSN 1362-3052 online
Ó 2010 Taylor & Francis
DOI: 10.1080/0951192X.2010.518976


sharing needs in product design and manufacture.
Consequently, there exists an ongoing requirement to
refine the understanding of the level of logical
expressiveness capable of semantically structuring the
meaning of product lifecycle concepts (Young et al.
2009, Chungoora 2010).
This article investigates the capture and intrasystem sharing of ontology-based knowledge using the
basis of two broad categories of ontological formalisms, notably ‘lightweight’ and ‘heavyweight’ approaches (Go´mez-Pe´rez et al. 2004), further explained
in the next section. By understanding the implications
of each approach applied to concepts in injection
moulding design and manufacture, the article contributes to a clarification of (1) the ways of expressively
capturing domain semantics and (2) the mechanisms
for sharing semantics across intra-system domains to
support engineering decisions.

A case study has been devised to expose the relative
strengths and weaknesses between a lightweight
ontological model and a version of the model
formalised using a heavyweight formalism. It has
been shown that the existence of an axiom layer in
the heavyweight model is paramount to capturing
rigorous semantics and for prompting the potential for
knowledge sharing. Moreover, certain characteristics


1060

N. Chungoora et al.

of the lightweight model have proved to be pertinent to
aiding intra-system knowledge sharing. Following this
case study, a suitable ontological direction is then
identified, as a benchmark for design and manufacture
domains that intend to exploit expressive semantics
alongside knowledge inference support.
2. Lightweight and heavyweight ontological
approaches
2.1.

A categorisation based on expressiveness

The requirements and preferences adopted by different
communities have led to the development and utilisation of various ontological formalisms. Ontological
formalisms are essentially formal languages that
support the construction of ontology-based models

and the encoding of the subject matter within these
models. Some commonly occurring formalisms are
illustrated in Figure 1, featuring the Unified Modelling
Language (UML 2009), frame-based languages (Wang
et al. 2006) and description logic-based languages
(Baader et al. 2007) among others. To distinguish
families of ontological formalisms, the ontology
community has introduced a categorisation based on
the expressiveness of the subject matter contained
within ontologies, and enabled via the use of ontological formalisms.
This categorisation involves the notions of ‘lightweight’ and ‘heavyweight’ ontological approaches,
which primarily differ in the degree of formality and
granularity with which they can represent the same
knowledge (Go´mez-Pe´rez et al. 2004, Casely-Hayford
2005). Lightweight models predominantly consist of a
taxonomy of concepts, with simple relationships
established among these concepts and very basic
constraints over the meaning of the ontological terms.
On the other hand, heavyweight models, in addition to

Figure 1. Examples of lightweight and heavyweight
ontological approaches.

having the lightweight structures, are accompanied by
a rich set of formal axioms that constrain the
interpretation of ontological terms. In Figure 1,
UML and Frames used on its own are examples of
lightweight ontological approaches (A), while DL and
frames with a first-order logic constraint language are
examples of heavyweight ontological approaches (B).

In the field of manufacturing engineering research,
both lightweight and heavyweight methods have been
used for the formalisation of domain models (ISO
18629 2005, Patil et al. 2005, Kim et al. 2006, Lin and
Harding 2007). It is clear, from the extent of the
exploited lightweight and heavyweight ontological
approaches, that there is currently no discernable
consensus on a preferred ontological direction. This is
largely because of the ongoing need to establish the
suitability of these approaches to meet the semantic
and knowledge sharing requirements of design and
manufacture. Hence, the aim of assessing the benefits
and limitations of both approaches becomes a key step
towards identifying the essential elements to progress
towards expressive ontology-based approaches.
2.2. Multi-domain knowledge representation and
sharing using lightweight and heavyweight approaches
The methodology to achieve the previously mentioned
aim is identified in Figure 2. Emphasis is placed on
multi-domain knowledge representation and intrasystem knowledge sharing in the context of injection
moulding. The methodology involves considering a
simple consumer product concept, namely a rotational
container (C), as shown in Figure 2. Using both
lightweight UML and heavyweight Frames with a firstorder logic constraint language, the product representation is first to be captured in the mouldability
domain (D), by using the semantic structures supported in both methods.
Then, populated knowledge from the mouldability
domain is to be shared with the mould design domain
(E) for obtaining a representation of the mould
product model knowledge. Following this stage, the
mould product model knowledge from the mould

design domain is to be shared with the mouldmanufacturing domain (F) to capture the manufacturing representation knowledge for the mould. The
sharing process between domains is to be achieved by
using the adequate translation/mapping mechanisms
accommodated in both ontology-based approaches.
A number of reasons justify the selection of UML
and Frames with a first-order logic constraint language, as the preferred lightweight and heavyweight
ontological formalisms respectively. In the first place, a
range of lightweight information models has exploited
UML for multi-viewpoint modelling applied to design


International Journal of Computer Integrated Manufacturing

Figure 2.

Methodology for multi-domain knowledge representation and sharing.

realisation stages (Tam et al. 2000, Kugathasan and
McMahon 2001, Canciglieri and Young 2009). Thus,
by performing an assessment of UML against the
exposed methodology, it becomes possible to provide
an appreciation of one extensively used lightweight
formalism.
Frames with a first-order logic constraint language
as heavyweight formalism presents characteristics that
overlap with a range of other heavyweight formalisms.
This explains its suitability for this investigation. For
example the formalism bears several structural similarities to description logic-based languages, which have
witnessed unprecedented relevance in product design
ontologies (AIM@SHAPE 2004, Lukibanov 2005).

Furthermore, the chosen heavyweight formalism holds
key commonalities with the common logic interchange
format (ISO/IEC 24707 2007), which has been used to
encode the process specification language (PSL)
ontology (ISO 18629 2005).
3.
3.1.

1061

Case study
Overview of case study

Figure 3 identifies the case study scenario for the
analysis of the selected lightweight and heavyweight
ontological approaches, to support multi-domain
knowledge representation and sharing. This scenario

provides a more detailed view on the use of the
methodology portrayed in Figure 2. Based on Figure 3,
the study concentrates on the analysis of a UML multidomain injection moulding model against a similar
model formalised in Frames with a first-order logic
constraint language. A detailed understanding behind
the UML development of the multiple viewpoint
domains can be found in an earlier manuscript
(Canciglieri and Young 2003).
For implementation purposes, the Knowledge
Engineering Methodology, prescribed by Noy and
McGuinness (2001), has been adopted during ontology
development. An appropriate UML tool has been

utilised to formalise the lightweight UML model.
Furthermore, the formalism Frames with the Prote´ge´
Axiom Language (PAL), as first-order logic constraint
language, has been used in the Prote´ge´ Frames 3.4
ontology editor (Prote´ge´ 2009) for representing the
heavyweight model.
An important facet of the case study is related to
the formal representation of the semantics of multidomain injection moulding and the corresponding
populated knowledge. This involves the following:
. the explicit representation of the mouldability,
mould design and mould manufacturing domain
semantics


1062

Figure 3.

N. Chungoora et al.

Case study scenario.

. representing knowledge that is either common
across domains or needs to be translated/mapped
to a different domain
. capturing logical pre-conditions that exist in one
domain that can drive the translation/mapping
of the appropriate knowledge.

manufacturing domain represents the semantics of the

rotational core insert from a machining viewpoint, for
example in terms of the types of machining features
that the rotational core insert holds in the manufacturing domain (K).

The representation of the rotational product in the
mouldability domain is partly comprised of internal
and external profiles (G) that pertain to primary and
transition features that form the wall of the product.
The dimensional knowledge captured in these profiles
is to be shared with the mould design domain (H). It is
to be noted that emphasis is laid on the semantics of
the internal profile of the rotational product, which are
then used to drive the rotational core insert representation knowledge (I) in the mould design domain. In this
case, the mould design domain has been referred to as
the ‘rotational core design domain’ to clarify that the
intended representation is for the rotational core
component of the mould.
The knowledge shared from the mouldability
domain to the rotational core design domain, is then
used to disseminate additional knowledge to the
rotational core-manufacturing domain (J) (i.e. the
mould manufacturing domain). The rotational core-

3.2. Lightweight ontology-based model
The lightweight UML model has been previously
documented (Canciglieri and Young 2009) and therefore,
this section concentrates on the most relevant strengths
and weaknesses carried by the lightweight UML
approach. Figure 4 provides a broad understanding of
the implementation of the lightweight ontology-based

model. In the model, UML class diagrams (L) have been
exploited to represent the necessary concepts and, to
some extent, the basic semantics of each domain. These
UML class representations capture domain concepts in
the form of classes and arrange these classes according to
a taxonomy. Relations with cardinality information (M)
are used to formulate the key associations that hold
between different classes.
The identification, retrieval and sharing of populated knowledge from the mouldability domain to the
rotational core design domain and from the design
domain to the manufacturing domain is formalised


International Journal of Computer Integrated Manufacturing

Figure 4.

1063

Using UML class and activity diagrams in the lightweight model.

through UML activity diagrams (N). These activity
diagrams enable the user to create a set of instructions
on how to translate the required attributes and
knowledge from one domain to another.
3.2.1. Strengths and weaknesses
. UML class diagrams provide a convenient way to
design ontologies, because they support a fairly
rich set of graphical constructs. This can be a
particularly useful means of reusing platformindependent ontologies prior to their implementation in the required ontology applications.

. The representation of multi-domain information
structures is dominated by the use of UML class
diagrams that involve taxonomies of classes and
cardinality relationships between classes, which
are fundamental to any ontology. From the
lightweight model explored, it has been possible
to exploit UML class diagrams to capture
common information content across domains.
. There are two main ways in which classes are
allowed to carry some semantics namely (1)
through the specification of traits that the classes
possess (i.e. attributes), and (2) by specifying
binary relations that hold between pairs of classes.


1064

N. Chungoora et al.

Classes, attributes and relations in UML hold
textual descriptions rather than semantic definitions. Consequently, domain concepts can only be
meaningfully interpreted if the implied semantics
of these concepts are understood by the user.
. UML activity diagrams allow translation/mapping knowledge to be captured and aid, at a system
development level, to automatically perform
information sharing procedures from one domain
to another. For example, it is possible to trigger the
automatic assertion of attributes and dimensional
knowledge from the mouldability domain to the

rotational core design domain. However, in the
experiment, because UML activity diagrams
depend on UML class diagrams, this implies that
translation procedures are dependent on the terms
carried by domain concepts rather than the
semantics of these concepts.
. Although a low level of computational interpretation can be captured in UML classes purely
associated to variations in class names, it is not
fully possible to embed pre-conditional knowledge and intent. For example in Figure 4, the
class name ‘Rot_Wall_Par_Part_Line’ (O) in the
mouldability domain is used to imply a rotational primary feature, which is positioned
parallel to a parting line configuration (P).
However, the condition for parallelism to a
parting line cannot be formally stated in UML.
3.3.

Heavyweight ontology-based model

The heavyweight ontological exploration using
Frames with the PAL differs both in the degree of
formality and granularity when compared with the
lightweight approach. Figure 5 depicts the heavyweight

Figure 5.

Heavyweight ontological structures.

ontological structures used to model multi-domain
semantics and to identify sharable knowledge between
domains.

In the heavyweight model, ontological structures
consist of taxonomies of classes, relations and functions, accompanied by a rigorous logic-based axiom
layer as shown in Figure 5 (Q). This layer is responsible
for supporting the meaning of concepts in computational form. The axiom layer is built on top of the basic
ontological structures and consists of integrity constraints and mapping rules, which are both written in
PAL. This constraint language accommodates firstorder semantics, thereby providing considerable flexibility in specifying the conditions for semantic
conformance and knowledge sharing. Integrity constraints are logical restrictions that help to ensure the
semantic integrity within the injection moulding
domains identified in Figure 3, while mapping rules
are logical conditions that help to identify potential
knowledge that could be shared from one domain to
the other.
Figure 6 provides a screen shot of the ‘mouldability
domain’ (R) class taxonomy in the class browser,
which at first glance is very similar to the class
taxonomy from the lightweight UML class model.
Other abstract classes are present namely ‘rotational
core design domain’ (S) and ‘rotational core manufacturing domain’ (T), which contain the information
structures for the rotational core insert design and
manufacture, respectively. The abstract class ‘Common Semantics’ (U) regroups reusable behaviours
across domains, such as the notions of ‘point’, ‘axis’,
‘length measure’ and ‘dimensional tolerance’ among
others.
An instance of the class ‘rotational mouldability
product’ (V) is shown in the instance browser.


International Journal of Computer Integrated Manufacturing
Captured semantics for one specific instance of
‘rotational mouldability product’, named ‘Product 1 Rotational Container’ (W), can be identified in

Figure 6. These semantic structures involve, for
example the list of point profiles aggregated under
the relations ‘holds_internal_profile’ (X) and ‘holds_external_profile’ (Y) and the list of primary and
transition features aggregated under the binary relation ‘holds_feature’ (Z).
3.3.1. Integrity constraints
From an ontology formalisation viewpoint, PAL is
used for model checking. This implies that integrity
constraints act as semantic prescriptions to ensure that
populated knowledge in the KB conforms to the
semantics expressed in the heavyweight model. To
verify whether asserted knowledge violates or conforms to semantics, integrity constraints can be
processed and a number of results are retained in the
event that these constraints have been infringed. In
other words, integrity constraints contribute to the
semantic integrity and enrichment of the KB.
To account for the semantic needs of the heavyweight model, integrity constraints have been written
for the multiple domains under consideration. Over 30
integrity constraints including both simple and complex ones have been modelled for all three domains.
The expression listed next gives an example of a simple
integrity constraint in the mouldability domain to

Figure 6.

Capturing the semantics of the mouldability domain.

1065

ensure that instances of the class ‘rotational mouldability product’ (see Figure 6 (V)) are only allowed to
hold one axis of rotation.
(defrange ?product :FRAME ‘Rotational Mouldability Product’)

(forall ?product
(¼ (number-of-slot-values holds_axis ?product) 1))
If, for example an instance of ‘Rotational Mouldability Product’ is asserted as having more than one
axis in the KB, then an execution of the PAL
constraint would show that this instance is violating
the fundamental semantics that a rotational mouldability product must always hold one axis. Figure 7
illustrates the result of querying an integrity constraint
based on an incorrectly populated knowledge element.
The instance ‘Product 1 - Rotational Container’ (W) is
shown to be violating the integrity constraint at query
time as a result of an additional ‘Probe Inconsistent
Axis’ (A1) having been assigned. The identification of
inconsistent knowledge, like the one shown in Figure 7,
provides a useful way of prompting the user to rectify
the incorrect assertions.
An example of a more complex integrity constraint
in the mouldability domain is shown in Figure 8. The
axiom captures the relevant logical pre-conditions to
ensure the correct specification of parting line features
(see Figure 4 (P)), by using the appropriate formalised
statement (B1). In the expression, the accurate


1066

N. Chungoora et al.

Figure 7.

Reporting an integrity constraint violation.

Figure 8.

Example of a complex integrity constraint.

definition of a ‘Parting Line Feature’ (C1) is captured
based on the known existence of some defined ‘Primary
Feature’ (D1). A similar understanding has been

followed for the specification of other simple and
complex integrity constraints required for the mould
design and mould-manufacturing domains.


International Journal of Computer Integrated Manufacturing

1067

The next stage following model checking of the
mouldability domain involves the execution of mapping rules. A first set of mapping rules is run with the
intention of identifying sharable knowledge that needs
to be communicated from the mouldability domain to
the rotational core design domain. Mapping rules are
written in the same way as integrity constraints and are
very similar in terms of complexity. The main
difference between the two lies in the specification of
existential quantifiers (i.e. the first-order logic directive
called ‘exist’) in the consequent of the mapping rules.
An example of a mapping rule is informally quoted
next, together with an exemplified understanding of the

implications of the mapping rule as shown in Figure 9.
‘Rotational core perpendicular straight line(s) must
be specified’ informally says that: If a parallel parting
line primary feature (e.g. (E1)) of a rotational
mouldability product has two points ?p1 and ?p2
that describe the internal profile of the product, such
that only the z coordinates of the two points are
different while the x and y coordinates are the same,
then there should exist a core perpendicular straight
line (e.g. (F1)) that meets the two points in the
rotational core design domain.

In simpler words, logical semantic conditions
arising in the mouldability domain imply the existence
of similar, modified or different knowledge elements in
the rotational core design domain. After sharable
knowledge is processed on running the first set of
mapping rules, the user then manually creates the
identified knowledge for the rotational core design
domain. Once this stage is performed, integrity
constraints for the rotational core design domain are
executed to ensure the consistency of the new knowledge input.
The next stage of knowledge sharing involves
discovering mappings from the rotational core design
domain to the rotational core-manufacturing domain.
An informally expressed example of a mapping rule in
this case is listed next, together with its corresponding
explanatory diagram in Figure 10.
‘Horizontal turning feature(s) must be specified’
informally says that: If a straight line (e.g. (G1)) that

defines the core insert for a rotational mouldability
product has two points ?p1 and ?p2, such that only the
x coordinates of the two points are different while the z
and y coordinates are the same, then there should exist
a horizontal turning feature profile (e.g. (H1)) that
meets the two points in the rotational core-manufacturing domain.
Results from similar mapping rules help identify
new knowledge required for the rotational coremanufacturing domain based on the knowledge
elements found in the rotational core design domain.
The user asserts the identified sharable knowledge in
the rotational core-manufacturing domain and ascertains that the knowledge input is consistent with
domain semantics by executing the integrity constraints for the manufacturing domain.
Figure 11 illustrates the results of processing two
mapping rules for each domain-to-domain sharable
knowledge identification process. Overall, 18 complex
mapping rules have been explored in the heavyweight

Figure 9. Example of a mapping rule for sharing between
the mouldability and rotational core design domains.

Figure 10. Example of a mapping rule for sharing between
the rotational core design and manufacturing domains.

Once the mouldability domain has been modelled
with its required instances populated, integrity constraints for that specific domain are processed to
ensure that the asserted knowledge concords with
semantics. If a constraint is violated this implies
that the query response, obtained from running
the constraint, points to inconsistent knowledge. The
consequence of a visible inconsistency prompts

the user to modify and/or assert correct knowledge.
The process of checking integrity constraints is iterated
until there are no violated conditions, thereby ensuring
the completeness of domain semantics.
3.3.2. Mapping rules


1068

N. Chungoora et al.
conditions deserve to be carefully written and
may result in an impedance in processing time as
a result of complexity, yet it is seen that such
constraints explicitly enable the subject matter of
the heavyweight model to be represented.
. In the heavyweight approach, the use of integrity
constraints has remained rigid. In other words,
integrity constraints can only be specified to
dictate the compulsory conformance of populated knowledge in the KB. In certain situations,
it could be necessary to also support optional
conformance of knowledge that is left for the
user to decide, as opposed to relying on the
system. This would require the ability to specify
integrity constraints of lesser ‘strength’, while
remaining traceable by the user.
. It is possible via the use of mapping rules to
make inferences for aiding the identification of
sharable knowledge that needs to be mapped
from one domain to another. However, in the
chosen heavyweight approach, it has not been

possible to automatically perform the assertion
of new knowledge as this has relied on manual
input, articulated through the processing of
mapping rules. This drawback is due to the fact
that PAL is essentially used for writing restrictions on existing knowledge rather than for
asserting new knowledge (Prote´ge´ 2009).
. Furthermore, the heavyweight ontology development process is absent of the use of suitable
ontology design schematics, which would serve
as platform-independent model. This is because
the ontology has been directly constructed within
the implementation environment. Hence, this
suggests that the heavyweight model is platform-dependent and, therefore, makes the process of interoperability between different
applications a potential issue.

experiment to obtain, from the product representation
in the mouldability domain, the accurate representation of the rotational core insert in both the mould
design and manufacturing domains.
In the sample results in Figure 11, it can be seen
that sharable knowledge has been inferred and consists
of (1) hole features from the mouldability domain that
can be directly shared with the rotational core design
domain (I1), (2) product geometry-related semantics
(J1) from the mouldability domain that are required in
the rotational core design domain and (3) geometryrelated semantics (K1) of the design rotational core
insert that are sharable with the rotational core
manufacturing domain, to obtain a machining feature
definition for the rotational core insert.
3.3.3. Strengths and weaknesses
. The fundamental, primarily lightweight, semantic structures of domain concepts can be readily
modelled through the specification of classes and

their taxonomies, accompanied by binary relations that hold between classes, and functions
that act like attributes of classes.
. The presence of an axiom layer provides the
capability to support the definition of rigorous
semantic structures, which complement the lightweight structures. The axiom layer accommodates a set of integrity constraints and mapping
rules written in the expressive and relatively
flexible PAL, which is first-order logic based.
. The axiom layer offers the ability to formally
capture logical pre-conditions. Although the
integrity constraints that model these pre-

4.

Figure 11. Samples of processed mapping rules for multidomain knowledge sharing.

Discussions

The case study has documented a set of strengths and
weaknesses of lightweight and heavyweight ontological
approaches, applied to multi-domain knowledge representation and sharing in injection moulding design
and manufacture. One of the primary differences
between the two approaches lies in the ability for the
heavyweight model to accommodate an axiom layer
supported by first-order logic semantics. The axiom
layer helps express the behaviours and conditions that
prescribe the integrity of populated domain knowledge
in KBs.
In UML 2, the object constraint language (OCL)
can be used to specify invariant conditions that must

International Journal of Computer Integrated Manufacturing
hold for the system being modelled (OCL 2006). This
implies that in a similar way to PAL, OCL would
enable the representation of certain logic-based conditions for ensuring that accurate knowledge is populated. However, OCL does not possess the expressive
power of first-order logic. For this reason, heavyweight
ontological approaches are favoured from the perspective of semantic expressiveness and interoperability.
Nevertheless, lightweight UML models can still be
effective in collaborative settings, provided the concepts defined in these models are agreed and
understood.
From the perspective of automating knowledge
assertion processes, UML activity diagrams hold a
stronger prospect of making the translation / mapping
of knowledge an easier task compared to mapping
rules. Thus, it can be extrapolated that by interfacing
UML activity diagram mechanisms with heavyweight
models, it could be possible to achieve an improved
method for automatically asserting new knowledge. In
the Prote´ge´ ontology environment, the Jess rule engine
reasoner (Prote´ge´ 2009) is able to interact directly with
populated knowledge and could, therefore, potentially
be exploited to perform the automatic assertion of
inferred knowledge from PAL mapping rules.
Additional opportunities exist for extending the
scope of the heavyweight ontology of injection
moulding design and manufacture. For example the
mould-manufacturing domain could be broadened to
include manufacturing process sequencing knowledge.
This is where, in particular, the PSL ontology would
help formalise the semantics of flow models (Bock and

Gruninger 2005). Such extensions would require a
heavyweight ontological approach that fully supports
more intricate relations and functions, as a result of the
complexity in the semantics of manufacturing process
sequences. For meeting this purpose, it would be
required to identify an even more expressive heavyweight ontological formalism, because using Frames
with a first-order logic constraint language in Prote´ge´
imposes certain restrictions to using binary relations
and less powerful semantic structures.
On the other hand, UML class diagrams as a
conceptual modelling method presents interesting
possibilities as far as platform-independent ontology
design is concerned. This is especially because currently, there is no de facto ontology design schematic
language. Therefore, for example, a UML class can be
used to represent a class in a heavyweight ontology and
a UML binary association can directly map to a binary
relation. Another example involves higher-arity relations, such as ternary relations, that can be represented
in UML by using the construct of n-ary associations.
However, more complex heavyweight constructs like
functions of multiple arities and the instantiation of

1069

meta-classes would demand an agreed mode of
exploiting UML class diagrams to avoid ambiguity.
5. Conclusions
The study presented in this work has shown that there
currently exist a number of benefits and drawbacks
related to both lightweight and heavyweight ontological approaches. It is understood from this that the
progress towards enabling expressive manufacturing

knowledge representation and sharing is bound to
enfold the integration of an array of ontology-based
understandings and semantic technologies.
Hence, the underpinning towards expressive ontology-based approaches firstly requires enabling multiple
domains to explicitly represent fundamental semantic
structures. These fundamental structures should include the notion of classes and their taxonomies,
together with relations and functions, which can bind
more than two classes together. At present, the
formalism Frames with a first-order logic constraint
language is not able to capture the representation of
these more complex relations and functions, while
UML also falls somehow short of a direct way for so
doing.
Second, integrity constraints need to be formalised
to complement these fundamental structures, thereby
semantically enriching ontologies and ensuring the
semantic consistency of populated knowledge. It is
highly desirable that integrity constraints be written
using an appropriate first-order logic-based language
to impart the required level of logical expressiveness.
Third, the process of ontology development should
be accompanied by the provision of appropriate
ontology design schematics. This is an essential stage
to support the platform-independent representation
and design of fundamental structures, prior to their
implementation. Currently, frames with a first-order
logic constraint language in Prote´ge´ is less suited for
the purpose of ontology design due to platformdependent modelling. Conversely, UML offers useful
prospects for ontology design, especially as it may be
possible to utilise UML in a customised way to

represent more complex ontological notions.
Finally, the ability to formalise knowledge inference rules, using a suitable first-order logic-based
language, should be provided as a means of deriving
new and expressive knowledge to support engineering
decisions. By providing a logic-based ground for
inference rules, it becomes possible to verify derived
knowledge via tractable reasoning procedures. These
procedures need to be accompanied by the relevant
translation/mapping mechanisms so as to help perform
the automatic assertion of derived multi-domain
knowledge.


1070

N. Chungoora et al.

With the introduction of new ontological formalisms, notably Common Logic (CL) (ISO/IEC 24707
2007), it is clear that improved capabilities are
emerging to address higher levels of semantic expressiveness and interoperability. Work is currently underway in our laboratory to take this aspect forward.
Acknowledgements
The research developed in this article has resulted from a
number of strands of work that have been supported by
different funding agencies. In particular, we wish to thank the
EPSRC, who are funding the majority of our work on
‘Interoperable Manufacturing Knowledge Systems’ (IMKS)
under project 253 of the Loughborough University Innovative Manufacturing and Construction Research Centre, and
the Wolfson School of Mechanical and Manufacturing
Engineering of Loughborough University for funding
research studentships.

References
AIM@SHAPE, 2004. Advanced and innovative models and
tools for the development of semantic-based systems for
handling, acquiring and processing knowledge embedded in
multidimensional digital objects. Available from: http://
www.aimatshape.net/ [Accessed March 2010].
Baader, F., Horrocks, I., and Sattler, U., 2007. Description
logics. In: F. van Harmelen, V. Lifschitz, and B. Porter,
eds. Handbook of knowledge representation. Amsterdam:
Elsevier, 135–179.
Bock, C. and Gruninger, M., 2005. PSL: a semantic domain
for flow models. Software and Systems Modelling
Journal, 4, 209–231.
Canciglieri, O.J. and Young, R.I.M., 2003. Information
sharing in multi-viewpoint injection moulding design and
manufacturing. International Journal of Production Research, 41, 1565–1586.
Canciglieri, O.J. and Young, R.I.M., 2009. Information
mapping across injection moulding design and manufacture domains. International Journal of Production Research. DOI: 10.1080/00207540902824974.
Casely-Hayford, L., 2005. Environments, methodologies and
languages for supporting users in building a chemical
ontology. MSc Dissertation. Faculty of Life Sciences,
University of Manchester.
Chungoora, N., 2010. A framework to support semantic
interoperability in product design and manufacture Ph.D.
Thesis. Loughborough University.
Go´mez-Pe´rez, A., Ferna´ndez-Lo´pez, M., and Corcho, O.,
2004. Ontological engineering: with examples from the
areas of knowledge management, e-commerce and the
semantic web. London, UK: Springer-Verlag.

ISO 18629, 2005. Industrial automation systems and integration – process specification language (PSL). Geneva,
Switzerland: International Organization for Standardization (ISO).
ISO/IEC 24707, 2007. Information technology – common logic
(cl): a framework for a family of logic-based languages.
Geneva, Switzerland: International Organization for
Standardization (ISO).
Kim, K.-Y., Manley, D.G., and Yang, H., 2006. Ontologybased assembly design and information sharing for
collaborative product development. Computer Aided
Design, 38, 1233–1250.
Kugathasan, P. and McMahon, C., 2001. Multiple viewpoint
models for automotive body-in-white design. International Journal of Production Research, 39 (8), 1698–1705.
Lin, H.K. and Harding, J.A., 2007. A manufacturing
engineering ontology model on the semantic web for
inter-enterprise collaboration. Computers in Industry, 58
(5), 428–437.
Lukibanov, O., 2005. Use of ontologies to support design
activities at DaimlerChrysler. Proceedings of the 8th
International Prote´ge´ Conference, Madrid, Spain.
Noy, N.F. and McGuinness, D.L., 2001. Ontology development 101: a guide to creating your first ontology.
Knowledge Systems Laboratory, Stanford University.
Object Constraint Language (OCL), 2006. OMG available
specification version 2.0. Available from: [Accessed
March 2010].
Patil, L., Dutta, D., and Sriram, R., 2005. Ontology-based
exchange of product data semantics. IEEE Transactions
on Automation Science and Engineering, 2 (3), 213–225.
Prote´ge´, 2009. Ontology editor and knowledge acquisition
system. Available from:
[Accessed November 2009].

Ray, S.R., 2004. NIST’s semantic approach to standards and
interoperability. PowerPoint presentation. Available
from: />NIST_Semantics–SteveRay_20040212a.ppt
[Accessed
November 2009].
Tam, S., et al., 2000. An object-based process planning and
scheduling model in a product design environment.
Logistics Information Management, 13 (4), 191–200.
Unified Modelling Language (UML). Available from: http://
www.uml.org/ [Accessed November 2009].
Wang, H.H., et al., 2006. Frames and OWL side by side.
Proceedings of the 9th International Prote´ge´ Conference,
Stanford, CA.
Young, R.I.M., et al., 2009. Enabling interoperable manufacturing knowledge sharing in PLM. In: Proceedings of
the 6th international product lifecycle management conference, University of Bath, Bath, UK.


International Journal of Computer Integrated Manufacturing
Vol. 23, No. 12, December 2010, 1071–1081

Automatic inspection of turbine blades using 5-axis coordinate measurement machine
Hui-Chin Changa* and Alan C. Linb
a

Department of Mechanical Engineering, De Lin Institute of Technology, No. 1, Lane 380, Ching-Yun Road, Tu-Cheng,
Taipei, Taiwan, Republic of China; bDepartment of Mechanical Engineering, National Taiwan University of Science and
Technology, 43 Keelung Road, Section 4 Taipei, Taiwan, Republic of China
(Received 25 October 2009; final version received 17 September 2010)
The complex geometric shape of turbine blades not only causes difficulties in fabrication but also makes it difficult to
examine the part’s precision when using a traditional 3-axis coordinate measurement machine (CMM). Generally

speaking, one has to use a multi-axis CMM and follow the appropriately planned measuring paths to accomplish the
task of precision measurement of the machined part. This article discusses the methodology of using a 3D geometric
model of a turbine blade as the basis to generate interference-free measuring paths that are suitable for implementing
on a CMM with three translational axes and two rotational axes. Through the use of the methodology proposed in
this research, the goal of multi-axis precision inspection for any geometric design of turbine blade can be achieved.
Keywords: automated inspection; CMM; collision-free; turbine blades; 5-axis measurement

1.

Introduction

Figure 1 shows the geometric shape of an axial wheel
with a pattern of turbine blade, which is one of the
indispensable components that are found in large
quantities in the aerospace and power industries. The
complex geometric shape and the limited space
between two neighbouring blades hinder the application of traditional techniques to inspect its precision.
In the process of measurement, an unexpected collision
caused by the measuring probe will break off the
measuring process or the measuring equipment itself to
be damaged. Therefore, the principle function of path
planning for coordinate measurement of a turbine
blade is to generate numerous orientations and
positions of the probe stylus, which are free of collision
during the measuring process. The traditional way of
planning measuring paths mostly relies on manual
editing and online coordinate measurement machine
(CMM) teaching, which is a time-consuming, highly
skilled and error-prone task. Generating an automated, collision-free measuring path thus becomes a
crucial issue for improving the quality of the measuring, as well as reducing the production lead-time for

high-precision turbine blades.
For prismatic parts with planar surfaces, the
concept of ‘ray tracing’ can be used to inspect the
collision between the start point and the target point of
the measuring path; if there is a collision, the algorithm
works through the topological structure of the part
and selects the midpoint of the edge, shared by the face

*Corresponding author. Email:
ISSN 0951-192X print/ISSN 1362-3052 online
Ó 2010 Taylor & Francis
DOI: 10.1080/0951192X.2010.527371


with which the path collides and the adjacent face
nearest to the target point, as the next probe point.
This procedure is followed till the target point is
reached (Lin and Murugappan 1998, 1999). The
method is valid only for planar surfaces but not for
sculptured surfaces. Furthermore, there is an approach
to using the geometric features set up beforehand to
generate measuring paths (Hermann 1997). This
approach can generate collision-free measuring paths
for objects with non-prismatic surfaces, such as
spherical surfaces, but for complicated sculpturedsurfaces, its feasibility cannot be verified. As for the
generation of collision-free measuring paths for
complicated sculptured-surfaces, Yau and Menq
(1995) and Menq and Lai (1992a,b) used 3D CAD
models of object and probe to determine whether an
intersection exists between them. This approach is

aimed at the inspection for objects with complicated
sculptured-surfaces, such as moulds and dies. Furthermore, various media, such as digital image data
(Takeuchi et al. 1990) and measured data points in
reverse engineering (Gupta and Sagar 1993), have been
studied in the past to plan appropriate paths for CMM
to undertake the measurement tasks. Nevertheless,
when it comes to the particular geometric shape of
turbine blades, in addition to their complicated
sculptured-surfaces, there are problems of overlap
and undercut between blades.
Chang and Lin (2005) using a 3-axis CMM
cooperate with a 2-axis automatic fixture to develop


1072

H.-C. Chang and A.C. Lin

the computer-aided measuring system, and the collision-free measuring path planning. Heo et al. (2008)
based on the ruled line information of a CAD database
for impeller blades, they projected hub and blade curve
on the XY plane, and to find the appropriate probe
postures according to the distribution of these projected curves that enable the probe to gauge the
inspection points without any collision between the
probe and the blade surfaces. In the meantime, they
partitioned the target blade surface into several
UMRs, which keeps the same probe approach vector
in each UMR so as not to change the probe
orientation. Then, the probe can be taught quite
simply in advance based on the ruled line information

of blade surface.
In addition, there are several studies of 5-axis NC
machining related to this topic, e.g. Balasubramaniam
et al. (2000, 2003) developed methods for 5-axis tool
positioning that account for accessibility of the tool
using visibility maps of the triangulated data though
the tool positions were not optimised for the highest
material removal rate. Gray et al. (2005) developed the
5-axis AIM, which is based on the fact that the widest
machined strip width is cut when the tool is tilted along
the feed direction. The concept of AIM is to tilt the
tool such that the tool axis remains in the tilting plane
and the forward bottom point of the tool remains in
tangential contact with the ccp (cutter contact point)
until a second contact point on the surface is touched.
This will give the smallest tilt angle and thus, the
largest effective radius resulting in the widest machined
strip width for the given feed direction at the ccp
without gouging the surface.
Considering the above problems as found in the
literature, this article addresses the methodology to
automate the planning of measuring paths of the 5-axis
CMM to suit the particular geometric shape of turbine
blades. It also discusses the implementation of the
proposed methodology to develop a computer-aided
measuring system for an automatic multi-axis precision inspection for turbine blades. The application of
the system significantly reduces the time spent on
manually calculating and editing the collision-free

Figure 1. Geometric shape of a workpiece with 19 pieces of

turbine blade.

measuring positions of the probe head, a bonus when
competing in the global marketplace.
2. Configuration of measurement
Besides the X-, Y-, and Z-axis, a typical 5-axis CMM is
equipped with a probe that provides motions in the Aand B-axis, as illustrated in Figure 2. In this article, the
orientation of the stylus is represented by (a, b), where
a and b are inclination angle and rotation angle,
respectively, and are defined as the rotational angles of
the stylus for the A- and B-axis. The initial orientation
of the A-axis of the stylus is set to be parallel to the
X-axis, i.e. the initial value of the inclination angle a is
908.
Every turbine blade in the rotor of a turbo-machine
presents an identical geometric shape. Therefore, it is
only necessary to take one blade for the planning of the
measuring paths. In this article, the Z-axis is set to be
along the rotational axis of the part, and the X–Y
plane is in the planar surface that slices the part, as
depicted in Figure 3.
3.

Establishing the criteria for interference avoidance

Two types of interference can be found in the
coordinate measurement of a turbine blade: interference between two blades and undercut interference
of a single blade. These interferences can be avoided
through proper adjustment of rotation angle b and
inclination angle a, respectively, as illustrated in

Figures 4 and 5. The methods are stated in the
following sections.
3.1. Avoiding the interference between two blades
Figure 6 shows relevant angles between the probe
stylus and turbine blades. In the figure, d is the angle

Figure 2.

Measuring probe of a 5-axis CMM.


International Journal of Computer Integrated Manufacturing
between two adjacent blades, f is the angle from the
X-axis to the line passing through the measuring point
P(Px, Py, Pz), and y is the angle that the current

1073

measuring point rotates around the Z-axis until
touching the next blade. The above three angles can
be formulated into the following equations:
d¼

360
n

ð1Þ

where n is the total number of blades.

Figure 3. Definition of X–Y plane and Z-axis for
measurement.

f ¼ tanÀ1 ðPy =Px Þ

ð2Þ

t
180
y ¼ d À qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Â
p
P2x þ P2y

ð3Þ

where t is the thickness of the cross-section at point
P.
Because a turbine blade is composed of pressure
surface and suction surface, as shown in Figure 1, in
calculating rotation angle b to avoid the collision
between the probe stylus and the turbine blades, the

Figure 4.

(A) Interference between two turbine blades, (B) adjustment of rotation angle b to avoid the interference.

Figure 5.

(A) Undercut interference of a single turbine blade, (B) adjustment of inclination angle a to avoid the interference.

1074

H.-C. Chang and A.C. Lin

surface that the measuring point locates has to be
taken into account. The equations are as follows:

Figures 6 and 7 illustrate the calculation of angle
b for a measuring point located on the pressure
surface, and Figures 8 and 9 are for a point in the
suction surface. In Equations (4) and (5), the

rotation of the stylus by angle b makes the stylus
locate on the middle of the safety boundaries, i.e. the
angle between the stylus and any of the two safety
boundaries is y/2, as shown in any of the above four
figures. This implies that the stylus has to be rotated
for every single point to be measured, whether a
collision happens or not. To minimise the total
number of stylus rotations while fulfilling the
requirement of free-of-collision, the stylus remains
not-rotated if the difference of angle b of the first
measuring point and the second one is smaller than
(1/n)6(y/2), where n ¼ 1.5 by default in this study,
and can be altered by the user.

Figure 6. Relevant angles for measuring a point on the
pressure surface and f 5 0.

Figure 8. Relevant angles for measuring a point on the
suction surface and f 5 0.

Figure 7. Relevant angles for measuring a point on the
pressure surface and f ! 0.

Figure 9. Relevant angles for measuring a point on the
suction surface and f ! 0.

for measuring points locate on the pressure surface:
b ¼ f þ y=2

ð4Þ

for measuring points locate on the suction surface:
b ¼ f À y=2:
ð5Þ


1075

International Journal of Computer Integrated Manufacturing
In a typical CMM, the minimal unit of change of
rotation angle of the stylus is 7.58; that is to say, the
angle of stylus orientation is always a multiple of 7.58.
If angle b is divided by 7.58, then the result can be
represented by an integer I and a decimal fraction R/
R
7.58, i.e. 7:5b  ¼ I 7:5
 . The formulae to determine angle b

are thus revised as follows:
If R < ð0:5 Â 7:5 Þ;
If R ! ð0:5 Â 7:5 Þ;

then

then

b ¼ I Â 7:5

b ¼ ðI þ 1Þ Â 7:5 : ð7Þ

The real height difference H(z) has to reflect the
diameter d of the ruby ball located at the tip of the
stylus:
HðzÞ ¼ Nz À Pz þ d=2

for pressure surface

ð10Þ

HðzÞ ¼ Nz À Pz À d=2

for suction surface

ð11Þ

ð8Þ

o ¼ tan À1 ðHðzÞ=Dðx; yÞÞ

Undercut area of a pressure surface.

ð12Þ

In other words, a is set to (refer to Figure 13):
a ¼ 90 þ o

ð13Þ

Since the minimal unit of angle change of stylus is
7.58, if angle o is not divisible by 7.5, any decimal
fraction is rounded to the integer.

Figure 11.

Top view of collision point N(x, y, z).

Figure 12.

Sectional view of undercut interference.

where Px is the x coordinate of measuring point P(x, y,
z); same with Py, Pz, Nx, Ny and Nz. Height difference

Figure 10.

ð9Þ

To avoid the occurrence of undercut interference, the

inclination angle a of the stylus has to be adjusted by:

The criteria developed in the earlier section for suction
and pressure surfaces only consider the interference in
the X–Y plane, without addressing the undercut
problem caused by the Z-directional twist of the
turbine blade. The front view in Figure 10 depicts an
undercut area of a suction surface.
During the probe-in process, undercut interference
occurs if the stylus touches any point on the silhouette
of the blade surface. Assuming that the initial
orientation of A-axis of the stylus is set to be parallel
to the X-axis, i.e. the initial value of inclination angle a
is 908. To acquire the collision point N(x, y, z) on the
silhouette, the rotation angle b of the stylus must be
found beforehand by employing the equations listed in
the earlier-mentioned section. With the current stylus
orientation (a, b), a ¼ 908, a straight line L(x, y, z) can
be drawn which passes through the measuring point
P(x, y, z) on the blade surface, the axis of revolution of
the turbine blade, and the silhouette of the blade
surface. Collision point N(x, y, z) is thus the point on
the silhouette, as illustrated in Figure 11. The
horizontal distance D(x, y) between points P(x, y, z)
and N(x, y, z) can be calculated by:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ðPx À Nx Þ2 þ ðPy À Ny Þ2

hðzÞ ¼ Nz À Pz

ð6Þ

3.2. Avoiding the undercut interference of a single
blade

Dðx; yÞ ¼

h(z) between the two points is (see Figure 12) as
follows:


1076

H.-C. Chang and A.C. Lin
Rotation angle b is calculated based on Equation (4):

3.3.

An example

The example shown in Figure 14 is used here to
illustrate how the above formulae work for the calculation of angles a and b. In the example, the total
number of turbine blades n is 15, the coordinate of the
measuring point P at the pressure surface is assumed to
be (18.67, 78.06, 78.39), the cross-section thickness t
at point P is 2 mm, and the diameter d of the ruby ball
is 2 mm. Equations (1)–(3) are used to find angles d, f
and y:
d¼

360 360
¼
¼ 24
n
15

f ¼ tan À1 ðPy =Px Þ ¼ tan À1 ðÀ8:06=18:67Þ ¼ À23:35
t
180
¼ 24
y ¼ d À qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Â
p
2
2
Px þ Py
4
À qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  57:3 ¼ 12:73
18:672 þ ðÀ8:06Þ2

b ¼ f þ y=2 ¼ À23:35 þ 12:73 =2 ¼ À16:99
By considering the minimal unit of change of
rotation angle, b is set as 7158, indicating the need for
rotating the stylus by this range of angle to prevent the
stylus from colliding with the blade that is adjacent to
the one currently under measurement.
Once angle b is determined, collision point N(x,
y, z) can be found by forming a straight line, which
passes through the Z-axis, the measuring point P and
the silhouette of the current blade. N(x, y, z) is the
point on the silhouette and its coordinate is found to

be (36.72, 715.21, 77.37). From Equation (8), the
horizontal distance D(x, y) between points P and N
can be calculated:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Dðx; yÞ ¼ ðPx À Nx Þ2 þ ðPy À Ny Þ2
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
¼ ½ð18:67 À 36:72Þ2 þ ðÀ8:06 þ 15:21Þ2 Š
¼ 19:41
The height difference H(z) is found by Equation (10):
HðzÞ ¼ Nz À Pz þ d=2 ¼ À7:37 þ 8:39 þ 2=2 ¼ 2:02
The positive value of H(z) implies that the height of the
collision point N is larger than that of the measuring
point P. In other words, undercut interference occurs
for the measurement of point P. To eliminate the
interference, the stylus is inclined by angle a and
Equations (12) and (13) are used to find its value:
a ¼ 90 þ o ¼ 90 þ tan À1 ðHðzÞ=Dðx; yÞÞ
¼ 90 þ tan À1 ð2:02=19:41Þ ¼ 95:94 :

Figure 13. Adjustment of inclination angle a to avoid
undercut interference.

Since the minimal unit of angle change of the stylus
is 7.58, angle a is set to be 97.58.
Figure 15 lists the final results of applying the
above procedure to calculate angles a and b for points
along the three rings of a turbine blade (20 points of
each ring).
4.

Figure 14.

An example for calculation of stylus orientation.

System implementation

The CMM adopted in this research to undertake the
measurement task of turbine blades is Mitutoyo
KN807 equipped with a measuring probe Renishaw
PH9A. The machine has three translational axes, X-,
Y-, and Z-axis, and two rotational axes, A- and B-axis,
which are able to rotate by 1058 and 1808, respectively,
at 7.58 per angle increment. A total of 720 stylus
orientations support the measurements of complex


International Journal of Computer Integrated Manufacturing

Figure 15.

1077

(A) Three rings of measuring points on a turbine blade, (B) angles b and a calculated for the measuring points.

surfaces. The control of the machine and its probe, as
well as the input and retrieval of the measuring data
are achieved by interface IEEE 488.
Regarding the development of a computer system
to implement the aforementioned methods for the
calculation of probe orientations, ACIS by Spatial

Technology, Inc., was used as the geometry kernel, and
Visual Cþþ was adopted as the programming
language. Detailed steps and some issues required to
be taken care of are listed below:
(1) First and foremost, the 3D CAD model of a
turbine blade is created using Pro/ENGINEER. Points to be measured are generated
on the part model, as pictured in Figures 15(a)
and 16. The coordinates, altogether with the
3D CAD model, become the input data into the
developed computer system. Normal vectors of
the measuring points are generated by the
system, and the resultant data are exported in
DMIS format so that the data can be used by a
typical CMM. Figure 17 shows a sample of the
exported data.
(2) The criteria setup in Section 3 are applied next
to find the stylus orientation (a, b) for every

Figure 16.
blade.

Three rings of measuring points on a turbine

measuring point. These data are then inserted
into the DMIS data.
(3) The DMIS data are converted into machine
codes acceptable by Mitutoyo KN807 by the
post processor developed in this research.
Figure 18 shows the dialogue box of the
DMIS data and CMM codes.

(4) Generate collision-free measuring paths and
insert them into the CMM codes.


1078

H.-C. Chang and A.C. Lin

Figure 17. Measuring points and normal vectors output in
DMIS data format.

Figure 19.

Travelling path during stylus angle change.

aiÀ1

Bz ¼ P0ðiÀ1Þz


PðiÀ1Þy
¼ tan À1
PðiÀ1Þx

Cðx; y; zÞ ¼ CðBx ; By ; diameter of turbine blade
Figure 18.

Sample of DMIS data and CMM control codes.

þ length of stylusÞ

B0x ¼ 1:5  turbine blade diameter  cos ai

If the change of stylus orientation is a prerequisite for the measurement of a specific point Pi,
then the change of stylus orientation should be
done away from the machined part, in order not
to cause a collision during the change of the
stylus orientation. As shown in Figure 19, when
the probe finishes the measurement of point
Pi71, which is a point before Pi, the probe
retracts to point A, moves from point A to B, lifts
from point B to C, changes stylus orientation at
point C, and then resumes its probing action.
The probe moves downwards from C to B0 (B0
denotes the new stylus orientation), and finally
reaches approach point P0i of point Pi. Coordinates of points B, B0 and C are calculated using
the following equations:
Bx ¼ 1:5  turbine blade diameter  cos aiÀ1
By ¼ 1:5  turbine blade diameter  sin aiÀ1

B0y ¼ 1:5  turbine blade diameter  sin ai
B0z ¼ P0iz
ai ¼ tan

À1



Piy
Pix



(5) Place the machined turbine blade on the CMM
table and conduct coordinate alignment.When
placing the part on the measurement machine
table, one will face the problem of aligning the
coordinate system of the CAD model, (Xcad,
Ycad, Zcad), with that of the measurement
machine (Xcmm, Ycmm, Zcmm). Figure 20 is a
typical method of coordinate positioning on a
CMM. First pick three points with the measuring probe to form an X–Y plane. Then take two
points to define the X-axis. Finally, obtain the
rotational axis of the part’s central hole to
place the Z-axis. The six degrees of freedom are


International Journal of Computer Integrated Manufacturing
thus fixed and the coordinate system is defined.
For the measurement of a turbine blade, the
X–Y plane and the Z-axis for measurement can
easily be defined by plane A and axis O on the
part, as depicted in Figure 21. However, the
X-axis of the part may pass through any point
on the blade surface, and thus the axis cannot
be defined in a straightforward manner. The
following steps are used to define the X-axis for
measurement: When placing the turbine blade,
observe in bare eye a minute angle between the
X-axis of the blade surface, Xcad, and the X-axis
of the CMM, Xcmm, as shown in Figure 22.

Randomly select a point Pcad from the CAD
model and record its coordinate C and normal
vector K. The CMM is operated manually
based on C and K, and the measured coordinate is, presumingly, Pcmm. After probe-radius
compensation, compare it with point Pcad to
obtain angle deviation s. This angle is used to
correct the coordinate system of the measurement process using the coordinate transformation function embedded in the CMM. Based on
the new coordinate system, the CMM is

1079

operated again manually, based on C and K.
The measured coordinate is again compared
with point Pcad, and the angle deviation is used
for coordinate correction. Repeat the same
procedure until the difference between Pcad and
Pcmm fits the expectations, as shown in Figure
23. If the difference between Pcmm and Pcad
cannot be effectively minimised, an alternative
that changes point Pcad and reinitiates X-axis
positioning might resolve the problem.
(6) Use the CMM codes generated in Step (4) to
conduct the actual coordinate measurement
task. Once the measurements of all points are
completed, conduct probe-radius compensation
by means of the points’ normal vectors. The
real turbine blade surface’s data can thus be
produced.
(7) Compare the data of compensated points with
the standard data that were obtained from the

3D CAD model, as shown in Figure 24, and the
error in between will appear. Figures 25–27 are,
respectively, the errors of the X-, Y- and Zcoordinate of the inner ring, middle ring and
outer ring.
5.

Conclusions

Turbine blades are widely used, and are especially
indispensable in the aerospace industry. But, the
complex surface geometry makes the inspection
difficult to execute. One usually must resort to a 5axis measuring instrument and spend a long time
manually planning the online measuring paths. This
article based on the geometry information of a CAD
database for turbine blades, and use simple basic
trigonometry to calculate and generate of 5-axis

Figure 20.
CMM.

Typical method of coordinate positioning on a

Figure 21. Positioning the XY-plane and Z-axis for
measurement.

Figure 22.

Initialising the X-axis for measurement.

1080

H.-C. Chang and A.C. Lin

Figure 27. Error of X-, Y- and Z-coordinate of the outer
ring.

Figure 23. Coordinate correction for positioning of the Xaxis.

Figure 24. Measured points and original contour of the
inner ring.

Figure 25. Error of X-, Y-, and Z-coordinate of the inner
ring.

Figure 26.
ring.

Error of X-, Y- and Z-coordinate of the middle

collision-free paths for the measurement of turbine
blades. The space between two neighbouring blades is
considered to find the proper rotation angle b for both
the suction and the pressure surfaces. On the other
hand, undercut interference of a single blade is avoided
by the adjustment of the inclination angle a. The
system implementation was carried out by using
Mitutoyo KN807 as the CMM, ACIS as the geometric
kernel and Visual Cþþ as the system development
tool. The system automatically generated collision-free

measuring paths in a short time. Successful implementation of measuring machined turbine blades verified
the viability of the proposed methodology for the
automatic planning of 5-axis measuring paths.

References
Balasubramaniam, M., Sarma, S., and Marciniak, K., 2003.
Collision-free finishing toolpaths from visibility data.
Computer-Aided Design, 35, 359–374.
Balasubramaniam, M., et al., 2000. Generating 5-axis NC
roughing paths directly from a tessellated representation.
Computer-Aided Design, 32 (4), 261–277.
Chang, H.C. and Lin, A.C., 2005. Automatic inspection of
turbine blades using a 3-axis CMM together with a 2-axis
dividing head. International Journal of Advanced Manufacturing Technology, 26 (7/8), 789–796.
Gray, P., Bedi, S., and Ismail, F., 2005. Arc-intersect method
for 5-axis tool positioning. Computer-Aided Design, 35
(7), 663–674.
Gupta, V.K. and Sagar, R., 1993. A PC-based system
integrating CMM and CAD for automated inspection
and reverse engineering. International Journal of Advanced Manufacturing Technology, 9, 306–310.
Hermann, G., 1997. Feature-based off-line programming of
coordinate measuring machines. In: INES Proceedings of
the IEEE international conference on intelligent engineering systems, 15–17 September 1997, Budapest, Hungary.
545–548.
Heo, E.Y., et al., 2008. Computer-aided measurement plan
for an impeller on a coordinate measurement machine
with a rotating and tilting probe. Robotics and ComputerIntegrated Manufacturing, 24, 788–795.
Lin, Y.J. and Murugappan, P., 1998. A new algorithm for
CAD-directed CMM dimensional inspection. Proceedings of the IEEE international conference on robotics and
automation, 16–20 May, Leuven, Belgium, 893–898.

International Journal of Computer Integrated Manufacturing
Lin, Y.J. and Murugappan, P., 1999. A new algorithm for
determining a collision-free path for a CMM probe.
Machine Tools and Manufacture, 39, 1397–1408.
Menq, C.H. and Lai, G.Y., 1992a. An intelligent planning
environment for automated dimensional inspection using
coordinate measuring machines. Journal of Engineering
for Industry, 114, 222–230.
Menq, C.H. and Lai, G.Y., 1992b. Automated precision
measurement of surface profile in CAD-directed inspection. IEEE transactions on robotics and automation, 8
(2), 268–278.

1081

Takeuchi, Y., Shimizu, H., and Mukai, I., 1990. Automatic
measurement of 3-dimensional coordinate measuring
machine by means of CAD and image data. Annals of
the CIRP, 39 (1), 565–568.
Yau, H.Y. and Menq, C.H., 1995. Automated CMM path
planning for dimensional inspection of dies and moulds
having complex surfaces. Machine Tools & Manufacture,
35 (6), 861–876.