Advanced Engineering Informatics 32 (2017) 92–112

Contents lists available at ScienceDirect

Advanced Engineering Informatics

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

Full length article

Research on a knowledge modelling methodology for fault diagnosis of

machine tools based on formal semantics
Zhou Qiang, Yan Ping ⇑, Xin Yang
State Key Laboratory of Mechanical Transmission, Chongqing University, Chongqing 400030, China

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

Article history: Fault diagnosis is a critical activity in PHM (Prognostics and Health Management) of machine tools due to
Received 2 September 2016 its great significance in such efforts as prolonging lifespan, improving production efficiency, and reducing
Received in revised form 7 December 2016 production costs. An efficient knowledge model is necessary to build an intelligent fault diagnosis system.
Accepted 7 January 2017
There have been several achievements in knowledge representation and modelling. However, due to their
various purposes and depths, the established knowledge models are less compatible, reusable or trans-
plantable, which restricts knowledge sharing and integration. A knowledge modelling methodology for
Keywords:
fault diagnosis of machine tools based on formal semantics (KMM-MTFD) is proposed in this paper to
Machine tool
Fault diagnosis
build an open, shared, and scalable ontology-based knowledge model of fault diagnosis of various
Knowledge modelling machine tools (OKM-MTFD). First, the proposed predicate-logic-based analysis method of fault elements
Ontology is adopted to study the fault diagnosis domain and extract the common domain knowledge, which
Formal semantics enables the establishment of the core ontology of OKM-MTFD to assure formal semantics. Next, using
the proposed two-stage classification method of fault elements and external ontology reference methods,
the core ontology can be extended into OKM-MTFD for a type or a specific machine tool. The knowledge
reasoning and querying methods based on OWL axioms, SWRL rules, special fault attributes and SPARQL
are provided to utilize the knowledge base efficiently. Finally, an ontology-based knowledge model and
knowledge base of a hobbing machine tool is presented to exemplify the validity of the proposed KMM-
MTFD.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction support customized nomenclature for different users and different

Machining systems that primarily consist of machine tools are
used in a wide range of industries. For example, machining in
China involves over 7 million machine tools [1]. However, due to
poor working conditions and long service time, the machine tools
are prone to breaking down and the same or similar faults tend
to occur repeatedly, leading to huge costs and time loss. The high
requirements and employment cost for fault maintenance person-
nel considered, a knowledge system able to meet the general
requirements for fault diagnosis is most interesting for manufac-
turing enterprises. As machine tools and fault types vary signifi-
cantly in different application contexts, even for the same type of
machine tool, machine tool manufacturers require a knowledge
system of fault diagnosis with great versatility to improve fault
diagnostic services and reduce operation costs. Moreover, the
knowledge system of fault diagnosis requires formal semantics to
⇑ Corresponding author.
E-mail address: labyp2016@163.com (P. Yan).
occasions, which contributes to promotion in practical applications
and acquisition of massive fault diagnosis knowledge for big data
analysis.
‘‘Smart Machine Tools” is an emerging topic of interest and the
focus of recent industry initiatives, and fault diagnosis is an impor-
tant part of machine health monitoring and an indispensable func-
tion of intelligent machine tools. With practical requirements and
the development trend of intelligent machine tools, a modelling
method based on knowledge for intelligent fault diagnosis has
become increasingly important. In recent years, research on
knowledge modelling has been developing rapidly [2]. The com-
monly used methods of knowledge representation are first-order
predicate logic [3], production rule [4] and semantic network and
framework [5]. Ontology can effectively combine the domain
knowledge with information representation and has many advan-
tages over other conceptual modelling technologies in classifica-
tion, sharing, and formalization [6–8].
Knowledge modelling of fault diagnosis based on ontology has
been paid much attention. Paper [9] conducted a comprehensive
review on fault diagnosis orientation and noted that ontology

http://dx.doi.org/10.1016/j.aei.2017.01.002
1474-0346/Ó 2017 Elsevier Ltd. All rights reserved.
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
Nomenclature
KMM-MTFD knowledge modelling methodology for fault diag-
nosis of machine tools
OKM-MTFD ontology-based knowledge model of fault diagnosis
of machine tools
FP fault phenomenon
FPA fault phenomenon attribute
engineering and hybrid methodologies have great potential. They
constructed an ontology-based knowledge-reasoning platform to
address and recognize fault propagation scenarios. Paper [10] pre-
sented a semantic knowledge model of fleets in the marine domain
based on ontologies for diagnostic purposes using a fleet-wide
approach, providing managers and engineers with more knowl-
edge and relevant and synthetized information regarding the sys-
tem behaviour. Paper [11] explicitly introduced FMEA (Failure
Mode and Effect Analysis) and used ‘‘can–cause” relations to corre-
late failure mode components, through which it presented a frame-
based method to construct an ontology model for fault diagnosis.
Paper [12] established a knowledge model of fault diagnosis for
wind turbines by extracting concepts in FMECA (Failure Mode,
Effects and Criticality Analysis). Though the paper studied the
structures of most mechanic equipment including machine tools,
commonality is still poor and the model is difficult to be modify,
restricting the portability for other equipment. Paper [13] studied
the concepts and relations of fault phenomenon, fault position,
fault cause and fault solution, established a knowledge model
based on ontology and constructed a knowledge base, which can
be reasoned to a certain degree. However, the maintainability
and scalability should be improved. Though ontology was adopted
as the modelling tool, the ability of semantic representation is lim-
ited because the ontology model is targeted to a specific machine
tool or purpose.
The existing ontology models of fault diagnosis can represent
the main concepts, classification, relations and some restrictions
according to specific machine tools’ characteristics and actual
needs. They can be rapidly modelled and easily carried out for
specified machine tools or machinery with simple structure and
few fault types. However, the problems listed below may restrict
their promotion for various machine tools.
 Shortage of semantic expression. As the concept of an ontology
class is too abstract and there are few limitations for syntax and
semantics, the semantics of knowledge cannot be identified or
checked by a reasoner for computers even if anything wrong
occurs. The expression may diverge or be informal, inaccurate
or even wrong due to different abilities, qualities and language
habits of different knowledge engineers.
 Specific machine oriented. The established fault diagnosis mod-
els are relatively fixed, which makes it difficult to add to, update
or/and delete any classes or relations. Once a model is changed,
its existing knowledge may no longer be valid. Thus mainte-
nance of knowledge is difficult, requiring quality staff and high
cost.
 Weak inference ability. The existing models are able to realize
reasoning to some degree, but only for existing knowledge.
They are unable to test the results or analyse new faults.
 Limited transplantation ability. The knowledge model should
be re-constructed for different machine tools to improve
reusability and commonality for transplantation and
promotion.
93
FM fault maintenance
FMA fault maintenance attribute
FC fault cause
FCA fault cause attribute
FL fault location
FAA fault auxiliary attribute
To address the aforementioned problems, this paper proposes a
knowledge modelling methodology for fault diagnosis of machine
tools based on formal semantics (KMM-MTFD) which provides uni-
fied domain ontology and methods for constructing an ontology-
based knowledge model of fault diagnosis with versatility, flexibil-
ity and reusability for various machine tools. Section 2 analyses the
development of fault diagnosis and knowledge modelling based on
ontology and discusses the requirements for a practical knowledge
system for fault diagnosis. Section 3 illustrates the concepts and
presents the principle of knowledge representation in this study,
and thus a novel knowledge modelling methodology for fault diag-
nosis KMM-MTFD including three layers, i.e., the core ontology
layer, knowledge layer, and reasoning and querying layer, is pro-
posed. Section 4 introduces the key methods and approaches in
detail: (1) predicate-logic-based analysis method of fault elements;
(2) core ontology of OKM-MTFD; (3) two-stage classification
method of fault elements; (4) external ontology reference method;
(5) construction approach of OKM-MTFD for a type of or a specific
machine tool; and (6) knowledge reasoning and querying methods.
Section 5 carries out an example for a hobbing machine tool using
KMM-MTFD. Comparisons and discussions are presented in Sec-
tion 6. Finally, the study’s conclusions are presented in Section 7.
2. Background and related works
2.1. Development of fault diagnosis
Fault diagnosis is a process of locating the exact causes of a fail-
ure or fault and finding its most likely solutions. Generally, fault
diagnosis technologies can be divided into three approaches, i.e.,
the artificial diagnosis method based on human experience, detec-
tion diagnosis method based on signal analysis and intelligent
diagnosis method based on knowledge processing. The artificial
diagnosis method based on human experience, which is heavily
dependent on the abilities and qualities of maintenance experts,
is achieved by evaluating the actual fault phenomenon and the
field environment with maintenance experts’ own experience
and knowledge. The detection diagnosis method based on signal
analysis detects probable fault locations using sensing and testing
technology, and then analyses possible fault causes by signal pro-
cessing, characteristics extraction and selection, pattern recogni-
tion and other data analysis technologies. Commonly used signal
processing methods include time and frequency domain analysis
[14], wavelet transform [15], EMD [16], etc.; commonly used pat-
tern recognition algorithms include artificial neural network [17],
support vector machine [18–20], Bayesian classification [21], fuzzy
theory [22,23], etc. Intelligent diagnosis method based on knowl-
edge processing, which is backed up with artificial intelligence
and knowledge engineering, finds existing or hidden knowledge
for fault diagnosis using inference engines. Representative meth-
ods include methods based on ontology models [11,24,25], expert
systems [26,27], fault tree analysis [28,29], case based reasoning
[30–32], etc.
94 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
The artificial diagnosis method based on human experience is
intuitive, fast, flexible and economical, but it is not immune to
problems such as high requirements for maintenance personnel,
low accuracy and weak diagnostic ability. The detection diagnosis
method based on signal analysis can be accurately, deeply and
thoroughly analyse the original data and find the profound causes,
especially for components or a single fault; however, it lacks inter-
pretation ability and, systemic and correlative analysis ability.
Because a machine tool is highly complex with numerous compo-
nents, its fault diagnosis requires specialized technical skills, rich
experience, and extensive understanding of the machine’s struc-
ture and operation and general concepts of diagnosis. Therefore,
the targeting knowledge system for fault diagnosis for machine
tools should be efficient in knowledge representation and organi-
zation to utilize the static fault knowledge adequately and effi-
ciently. It should also have the ability to combine the above two
diagnosis methods to facilitate accurate and deep analysis in micro
respect and quick systematic diagnosis in macro respect.
2.2. Knowledge modelling based on ontology
Paper [33] surveyed publications relating to knowledge mod-
elling and manipulation technologies from the year 2000 to 2015
and classified them into 4 categories: linguistic knowledge-based,
expert knowledge-based, ontology and cognitive knowledge-
based. Among them, ontology engineering has made a significant
contribution to advancing engineering tools and software for the
development of knowledge modelling and thus has proved its
great effectiveness in various applications including fault diagnosis
[34].
The concept of ontology originates from philosophy, and it was
later introduced into computer science and artificial intelligence.
Gruber defined ontology as an explicit specification of a conceptu-
alization [35]. Ontology can explicitly express various concepts and
interrelations in a specific field, so it plays an important role in
semantics. Because ontology adopts precise formal language, pre-
cise syntax and clear semantics, the relations between concepts
and objects in the problem domain tend to be clarified, which sig-
nificantly reduces possible misunderstandings of the concepts and
logic relations in the problems [36].
OWL [37] is the standard ontology language recommended by
W3C. It facilitates machine interpretability of contents by provid-
ing additional vocabulary and formal semantics. OWL allows only
binary relations between classes. However, rules in the fault diag-
nosis domain always contain many objects and thus go beyond the
ability of OWL. SWRL [38], based on OWL, is used to represent rules
and extend the expressivity of OWL and its set of axioms to include
Horn-like rules [2]. The combination of both languages creates a
more powerful ontology language in knowledge representation
and reasoning capabilities [39]. Thus, this paper will use OWL as
the ontological modelling language and SWRL as the rule language.
2.3. Fault diagnosis based on knowledge
Through learning and utilizing the existing knowledge of fault
diagnosis, the intelligent fault diagnosis based on knowledge aims
to find the unknown knowledge and handle new and complex sit-
uations. There are two main factors affecting performance: the
amount of domain knowledge and the ability to apply that knowl-
edge to solve actual problems.
According to the selection of domain knowledge, the knowledge
used in intelligent fault diagnosis can be divided into two classes:
shallow knowledge and deep knowledge [40]. The former indicates
the experiential knowledge of domain experts, and the latter is the
knowledge regarding the structure, function or basic principle of
diagnostic objects. Papers [28,29] constructed a fault tree with
shallow knowledge of expert experience, and applied it to the fault
diagnosis of a gearbox. Paper [41] summarized the most commonly
used algorithms in the PHM (Prognostics and Health Management)
field, including their applications, advantages and weaknesses for
selecting appropriate analytical tools according to various types
of data, which contributed to deep knowledge in processing and
analysis original fault data. Papers [42,43] diagnosed the faults of
rotating machines using shallow knowledge in the form of rules
and adopted the EMD method to preprocess the vibration signals
to mine the fault characteristic information more accurately, which
can be seen as an application of deep knowledge in signal analysis.
Shallow knowledge has advantages such as intuitive expression
of knowledge, unified format and quick reasoning and disadvan-
tages such as difficulty in constructing comprehensive knowledge
sets, overflow beyond existing knowledge and competition or con-
tradiction between rules. Deep knowledge can reflect the internal
laws or relations buried in the diagnosed objects with support of
multiple disciplines, but it has difficulties establishing a practical
model with efficient inference ability. Currently, first-generation
expert systems based on shallow knowledge and second-
generation expert systems based on deep knowledge move for-
ward with a hybrid structure [27,44].
However, to implement a practical knowledge system for fault
diagnosis, the following problems and requirements should be well
considered in practical applications. First, an efficient knowledge
model for representing, organizing and utilizing massive fault
knowledge should be established and it should be updatable,
extensible and adaptable for various types of machine tools. Sec-
ond, shallow knowledge and deep knowledge can be integrated
without increasing the complexity of the model or reasoning
mechanism dramatically. Third, knowledge inference and querying
must be well-supported to use the knowledge base flexibly to find
the hidden knowledge.
3. Overview of KMM-MTFD
3.1. Concepts in fault diagnosis of machine tools
Various machine tools, complicated systems consisting of
mechanical, electrical, hydraulic and other subsystems, have
numerous parts and are thus challenging for fault diagnosis. Faults
of machine tools come from many aspects such as installation,
operation and maintenance and there are many complex charac-
teristics of machine tools such as irregularity, uncertainty and cor-
relation. Numerous classification methods of machine tool faults
have been applied, which reflects the characteristics of faults from
different research perspectives [45].
FMECA (Failure Mode, Effects and Criticality Analysis) [46–48]
is one of the earliest systemic techniques for failure analysis and
is utilized to analyse the reliability and safety of equipment with
inductive logic. FMECA is typically used to assess the impact of
potential faults on equipment functions, reliability, maintainabil-
ity, maintenance personnel and the environment, and hence to
prevent, or minimize the criticality of equipment faults. This paper
will adopt FMECA as a knowledge source for fault diagnosis, as
FMECA contains almost all the information of faults regarding
equipment components [49].
Based on the analysis of the current research, FMECA, fault diag-
nosis and maintenance records from machine tool manufacturers,
the main research contents of fault diagnosis of machine tools
are extracted as shown in Table 1.
According to current researches, there are four key concepts in
the fault diagnosis domain, i.e., fault phenomenon, fault mainte-
nance, fault cause and fault location. This paper defines them as
a ‘‘fault element”, as shown in Def 1. The above research shows
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
Table 1
Main research contents of the fault diagnosis domain of machine tools.
No. Perspective Classification
1 Fault Scratch/Leakage/Damage/Noise/. . .
Phenomenon
2 Fault Location Subsystem-Mechanic/Electric/Hydraulic/
Pneumatic/. . .
3 Fault Cause Cause-Mechanic/Electric/Operational/
Environmental/. . .
4 Frequency Sudden/Occasional/Frequent. . .
5 Severity Slight/General/Serious. . .
6 Alarm functions No alarm/Hardware alarm/Software alarm. . .
7 Fault Type Systematic/Random/. . .
8 Detect method Manual/Controller/Sensor. . .
9 Maintenance Repair/Repack/Adjust/Reset/. . .
10 ... ...
that fault diagnosis of machine tools involves a wide range of
research perspectives, as presented in Table 1. As the requirements
and the characteristics of fault elements vary, description and rep-
resentation will be carried out in different ways to meet various
demands. For example, machine tool manufacturers are concerned
with the causes of faults while the users care more about the sever-
ity and the breakdown time. From the perspective of commonality,
this paper defines a ‘‘fault element attribute” for each fault element
as shown in Def 2. Therefore, the information needs can be satis-
fied separately. The classifications listed in Table 1 can be defined
as fault element attributes, such as fault cause, fault location,
severity, detect method, frequency, etc. Specifically, as the location
of the FP, FM or FC may not be the same, the FL will be treated as
their attribute. To be noticed, as FL will be mostly imported from
the ontology model of structure, its attribute will not be defined
in this paper.
Def 1 (Fault element). The basic units of machine tool fault
diagnosis (MTFD), including fault phenomenon (FP), fault mainte-
nance (FM), fault cause (FC), fault location (FL), etc.
Def 2 (Fault element attribute). Property or feature of a fault ele-
ment, including fault phenomenon attribute (FPA), fault mainte-
nance attribute (FMA) and fault cause attribute (FCA).
With the mechanism of fault elements and fault element attri-
butes, the concepts can be clearly defined on demand. For example,
for a typical FP of ‘‘the running state of spindle is not stable”, the
fault location FPA is ‘‘spindle”, the severity FPA is ‘‘general”, the fre-
quency FPA is ‘‘rarely”, etc. The attributes can be updated, added to
or deleted according to the latest situation without changing the
established relations among fault elements. Moreover, the new
hidden relations can be reasoned out with inference engines,
ensuring dynamic maintainability of the knowledge base. The
common attributes such as severity, frequency and emergency
can be reused for newly added fault elements, reducing knowledge
redundancy.
For a more accurate description of fault elements, fault element
attributes are further divided into fault element characteristic
attribute, fault element intrinsic characteristic attribute and fault
element general attribute, which is the basis of the detailed
description of fault elements and advanced knowledge utilization
such as the proposed two-stage classification method in Sec-
tion 4.3. The definitions are listed as Defs 3–5. Their relations are
shown in Fig. 1. Take fault phenomenon as an example, the intrin-
sic characteristic attribute reflects the essence of the FP and tells
what it is and each FP must have one and only one intrinsic char-
acteristic attribute. The characteristic attribute groups similar FPs
with common attributes. Therefore, the FPs can be initially anal-
95
Fault Element Attribute
Characteristic Intrinsic General
Attribute Attribute Attribute
Fig. 1. Relationship of fault element attribute.
ysed and the consistency will be checked immediately. The general
attribute defines FP more delicately. The attributes in the example
FP of ‘‘the running state of spindle is not stable” can thus be further
divided. The characteristic attribute and the intrinsic characteristic
attribute is the same, i.e., ‘‘not stable”; the general attributes are
fault location, severity and frequency as mentioned above. For
another example FP of ‘‘Y axis movement is not stable while the
ball screw is loose”, ‘‘not stable” is its characteristic attribute, while
the intrinsic characteristic attribute is ‘‘loose” because it is closer to
its essence.
Def 3 (Fault element characteristic attribute). The fault element
attribute directly relates to the fault element itself, or character-
istic attribute for short. It includes fault phenomenon characteristic
attribute, fault maintenance characteristic attribute and fault cause
characteristic attribute.
Def 4 (Fault element intrinsic characteristic attribute). The fault ele-
ment attribute that can only reflect the intrinsic characteristic of
fault element, a subset of characteristic attribute, or intrinsic attri-
bute for short. It includes fault phenomenon intrinsic attribute and
fault maintenance intrinsic attribute.
Def 5 (Fault element general attribute). The fault element attribute
that reflects the property of fault element from different perspec-
tives, the complement of characteristic attribute, or general attri-
bute for short. It includes fault phenomenon general attribute,
fault maintenance general attribute and fault cause general
attribute.
3.2. Knowledge representation based on semantic network
To illustrate the concepts and relations presented above in
detail, this section introduces a knowledge representation method
based on a semantic network, which is the basis of KMM-MTFD. As
shown in Fig. 2, concepts (including the main concept and related
concepts in a certain domain) of fault elements are expressed using
nodes and their semantic relations are expressed via arcs. In ontol-
ogy and OWL, fault elements and fault element attributes can be
defined as ontology classes and the relations can be defined as
object property and data property. Thus, the semantic network
with formal semantics can be constructed by virtue of OWL
axioms. The core ontology of main concepts in the fault diagnosis
domain is constructed in Section 4. Moreover, several methods
are proposed for applying and utilizing semantic-network-based
knowledge more efficiently.
After reaching a certain scale, the newly added elements such as
classes or instances will be connected to the semantic network as
nodes with specific semantics according to their attributes and ini-
tial relations with existing nodes and the hidden semantics can be
found with inference engines. The ontology element, namely, class,
object property and data property can be compatible or integrated
with other similar domain ontologies using ontology calibration or
ontology mapping technology. In addition, OWL supports various
96 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
Fault Node: center concept
FP Node: Directly related concept
FPA1 Node: Indirectly related concept
FC1
Arc: relations
FPA2
FP
FPA1
FP Fault Phenomenon FM Fault Maintenance
FPA Fault Maintenance Attribute FMA Fault Maintenance Attribute
Fig. 2. Semantic network of the fault diagnosis domain.
names, alias or languages for one concept with the same meaning
using self-defined labels, ensuring semantic consistency and envi-
ronmental adaptability. For example, class FP can also be named as
fault phenomenon, failure phenomena, trouble phenomena, etc.
From the perspective of knowledge representation, concepts in
the fault diagnosis domain are defined separately in the semantic
network based method, which can be easily integrated with the
existing domain knowledge, including shallow knowledge and
deep knowledge. For example, FPs of a single bear can be identified
using feature extraction and state identification method for its
vibration signal, supporting intelligent fault diagnosis from sensors
to fault maintenance method. For FM, the related knowledge such
as fault alarms, maintenance personnel assignment and cost
accounting can be integrated conveniently to improve automatic-
ity and intelligence.
3.3. KMM-MTFD
In a review of related works, the existing modelling methods,
which are dedicated to specific purposes or specific machine tools,
make it difficult to present an overall description of fault elements.
They lack effective extension mechanisms conducive to building an
intelligent fault diagnosis system. Therefore, it is necessary to
establish a knowledge model of fault diagnosis that is also not
comprehensive for various aspects of fault diagnosis but can also
be popularized to a specific machine tool.
A knowledge modelling methodology for fault diagnosis of
machine tools (KMM-MTFD) is put forward in this paper, which
provides a common domain ontology and a set of supporting
methods and techniques for constructing an ontology-based
knowledge model of fault diagnosis of machine tools (OKM-
MTFD). KMM-MTFD includes the acquisition, analysis and repre-
sentation of fault diagnosis knowledge, the core ontology of
OKM-MTFD, the construction approach for a type of or a specific
machine tool and the reasoning and querying methods. KMM-
MTFD is organized in a hierarchy of three layers as shown in
FC1
FC
FCA
FM
Fault
FMA
FL
FL1 FL2
FC Fault Cause FL Fault Location
FCA Fault Cause Attribute FAA Fault Auxiliary Attribute
Fig. 3: the core ontology layer, knowledge layer, and reasoning
and querying layer.
(1) Core ontology layer
This layer adopts the predicate-logic-based analysis method of
fault elements proposed in Section 4.1 to conduct a comprehen-
sive, detailed and complete analysis and description of fault ele-
ments and provide formal semantics for knowledge
representation. Based on analysis of a large number of fault diag-
nosis records, the main concepts, relations and hierarchical struc-
ture are extracted for establishment of the core ontology of the
fault diagnosis domain, which provides a common ontology for
constructing an OKM-MTFD for a specific machine tool. Moreover,
the integration ability and consistency of the constructed knowl-
edge models and knowledge bases shall be assured by taking the
core ontology of OKM-MTFD as a common domain ontology.
(2) Knowledge layer
The knowledge layer constructs an ontology-based knowledge
model for a specific machine tool following a general construction
approach for ontology modelling by extending the core ontology of
OKM-MTFD using the two-stage classification method of fault ele-
ments and the external ontology reference method. The two-stage
classification method of fault elements enables defining the
classes, objects properties and relations of fault elements for speci-
fic characteristics (such as machine structure, fault collection and
monitoring method and fault level of emergency) of a particular
machine tool to be defined. The external ontology reference
method can integrate the existing ontology directly, which
improves reusability and reduces dependence on a specific
machine tool. Finally, combined with the knowledge table of fault
diagnosis, the knowledge base based on OKM-MTFD can be con-
structed. Additionally, the knowledge model and knowledge base
can be maintained dynamically and effectively.
(3) Reasoning and querying layer
This layer studies the knowledge reasoning and querying prob-
lem for the knowledge base based on the core ontology of OKM-
MTFD. By virtue of an inference engine, this layer can carry out
knowledge reasoning of fault diagnosis, analysis of newly added
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
Fig. 3. Knowledge modelling methodology for fault diagnosis of machine tools.
FPs and treatment of special fault problems using the knowledge
reasoning methods based on OWL (Web Ontology Language)
axioms, SWRL (Semantics Web Rule Language) rules and special
fault attributes respectively. A querying method based on SPARQL
(SPARQL Protocol and RDF Query Language) [50] is proposed for
four types of knowledge query problems, i.e., querying FM from
FP, querying FP and FM from FC, querying FP, FM and FC from FL
and querying FP from FM. The proposed reasoning and query
methods provide support for efficient utilization of the knowledge
base and improvement of intelligence fault diagnosis.
4. Methods and approaches in KMM-MTFD
4.1. Predicate-logic-based analysis method of fault elements
A predicate-logic-based analysis method of fault elements is
used to analyse linguistic components of fault elements to acquire
a set of standard terminologies for fault diagnosis of machine tools,
assuring unified formal semantics. By adopting the predicate logic,
the method can also solve the expression diversity problem exist-
ing in current researches and thus enhance semantics in fault ele-
ment description. Furthermore, every linguistic component is
abundant in semantics and can be associated with the others
through appropriate predicates to support the creation of new
97
knowledge that can be edited, reused and maintained according
to practical needs, providing basic elements and logical support
for intelligent fault diagnosis. The predicate-logic-based analysis
method of fault elements is carried out as shown in Fig. 4.
First, fault elements such as FP, FM, FC and FL should be sepa-
rated from the original fault diagnosis records. Second, properties
of every fault element should be extracted in the ‘‘Subject-Verb-
Object” form, in which the subject and object reflect the body or
property and the verb reflects their relations. Third, each property
formatted in ‘‘Subject-Verb-Object” can be seen as an atomic
proposition and can be further divided into individual terms, pred-
icates, quantifiers, etc. The process should be carried out for all
properties and fault elements. Finally, after analysis of a large
number of fault records, libraries of individual terms, predicates
and quantifiers are obtained.
Consequently, these libraries are analysed and summarized to
build a library of terminologies for fault diagnosis of machine tools.
The individual terms, as subject or object, can be refined to classes
in ontology, such as fault element classes and fault element attri-
bute classes; the predicates can be refined to object properties in
ontology; quantifiers and the others can be refined to data proper-
ties, cardinality or restrictions in ontology. For analysis of massive
fault diagnosis records, rough set theory can be utilized for analysis
and extraction of fault elements to construct ontology, which will
be studied in-depth in the following research.
98 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
Original
fault diagnosis records
Fault elements
FP,FM,FC,...
Fault properties
In
Subject-Verb-Object
Individual terms
Predicates
Quantifiers
...
Fig. 4. The predicate-logic-based analysis method of fault elements.
The relations between fault elements and fault element
attributes are defined and described by using a predicate calcu-
lus formula, providing predicate logic support for building
intrinsic class, functional class, attribute class and object prop-
erty in the following sections. The Manchester OWL syntax [51]
borrows ideas from the OWL Abstract Syntax and supports the
predicate logic, but it is far less verbose. Therefore, this
paper adopts Manchester OWL syntax to present the predicate
calculus formula and the axioms, relations and rules in
ontology.
4.2. Core ontology of OKM-MTFD
Based on the analysis of the original fault diagnosis records and
the research contents of the fault diagnosis domain of machine
tools listed in Table 1, the fault diagnosis problem of machine tools
is decomposed into fault elements and fault element attributes.
Based on the analysis of the characteristics and relationships, the
classes, object properties and hierarchy of the core ontology of
OKM-MTFD are established.
4.2.1. Classes of the core ontology of OKM-MTFD
This paper uses ‘‘MTFault” as the top ontology of the fault diag-
nosis domain of machine tools. Based on the above research, eight
Upper Level Classes (ULCs) are established to cover the main con-
cepts in the fault diagnosis domain of machine tools. Fig. 5 illus-
trates the ULCs and their relationships, including
ZFaultPhenomenon, ZFaultMaintenance, ZFaultCause, ZFaultLocation,
ZFaultPhenomenonAttribute, ZFaultMaintenanceAttribute, ZFaultCau-
seAttribute and ZFaultAuxiliaryAttribute.
The ULCs of the core ontology of OKM-MTFD and their sub-
classes are defined as follows:
(1) Fault phenomenon refers to the phenomenon that the
machine tools cannot work properly during operation. FPA
reflects the properties of FP from different perspectives
according to demands, including the location, cause, type,
frequency, severity, alarm types, detection methods, etc.
The ontology class is defined as follows:
Ontology
fault element class
fault element attribute class
object property
data property
restrictions
...
 FP is represented by class ZFaultPhenomenon, including
the intrinsic class of FP ZFaultPheIntrinsic and functional
class of FP ZFaultPheFunctional;
 FPA is represented by class ZFaultPhenomenonAttribute,
including subclasses such as ZFaultPheAttrCharacteristic,
ZFaultPheAttrDegree, ZFaultPheAttrFreq, etc. ZFaultPheno-
menonAttribute can be updated according to specific
machine tools.
(2) FM includes preventive maintenance and breakdown main-
tenance. FMA reflects the properties of FM from different
perspectives according to demands, including location,
cause, type, frequency, severity, emergency degree, cost,
etc. The ontology class is defined as follows:
 FM is represented by class ZFaultMaintenance, including
the intrinsic class of FM ZFaultMntIntrinsic and functional
class of FM ZFaultMntFunctional;
 FMA is represented by class ZFaultMaintenanceAttribute,
including subclasses like ZFaultMntAttrType,
ZFaultMntAttrEmergencyDegree ZFaultMntAttrCost, etc.
ZFaultMaintenanceAttributecan be updated according to
specific machine tools.
(3) FC refers to the cause behind the machine tools not working
properly during operation, including mechanical, electrical,
hydraulic, operational, environmental, etc. The FCA reflects
the properties of FC from different perspectives, including
the location, type, frequency, severity, etc. As most FP classes
are related to FM classes through FC classes, the description
level of details of FC has important influence on the validity
of fault reasoning, depth of fault diagnosis, accuracy for
guiding fault maintenance. The ontology class is defined as:
 FC is represented by class ZFaultCause, including sub-
classes like ZFaultCauMechanic, ZFaultCauElectric, ZFault-
CauEnv, etc.;
 FCA is represented by class ZFaultCauseAttribute, includ-
ing subclasses like ZFaultCauAttrFrequency, ZFaultCauAttr-
Degree, ZFaultCauAttrType, etc.
(4) FL refers to the location where the FP, FM, FC happened or
related. FL is related to the structure of specific machine
tools and the description level of details of which has impor-
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112 99

MTStatus MTFaultStatus
(exernal)
SubClassOf
(exernal)
REF MTFault

MTStructure REF ZFaultLocation SubClassOf SubClassOf ZFaultCause

(exernal)

AuxiliaryAttribute
(exernal)
REF ZFaultAuxiliaryAttribute SubClassOf SubClassOf ZFaultCauseAttribute

ZFaultPhenomenon SubClassOf SubClassOf ZFaultMaintenance

ZFaultPhenomenonAttribute SubClassOf SubClassOf ZFaultMaintenanceAttribute

Fig. 5. The Upper Level Classes of the core ontology of OKM-MTFD.

tant influence on the depth of fault diagnosis. To assure the
commonality of OKM-MTFD, it is recommended to refer
external structure ontology for specific machine tools. The
ontology class is defined as ZFaultLocation.
(5) The auxiliary fault attribute is used to describe the auxiliary
element in fault diagnosis records, including the behaviour,
action, status in CNC and adjectives and adverbs in linguis-
tics to enhance the descriptive ability. It is usually referred
to as an interface for external ontology and support for
OKM-MTFD. The ontology class is defined as
ZFaultAuxiliaryAttribute.
Table 2 lists the main classes of OKM-MTFD, including the fault
elements and fault element attributes. The intrinsic and functional
classes are introduced in Section 4.3. The classes of fault element
attributes should be adapted, extended and modified for specific
machine tools and requirements.
4.2.2. Properties of the core ontology of OKM-MTFD
Property is used to describe the binary relationship between
individuals, including object property and data property [10]. The
attributes of fault elements and their relationships can be seman-
tically defined by appropriate and sufficient object properties. For
FP, each class or individual has only one FP intrinsic attribute
and various FPAs from different perspectives. The FP is expressed
as ‘‘intrinsic attribute + multi-attributes of FP” by means of object
property. Similarly, FM is expressed as ‘‘intrinsic attribute
+ multi-attributes of FM” by means of object property. Fig. 6 illus-
trates the main relationships between fault elements and their
own attributes and gives some object properties.
This part defines the main object property of the core ontology
of OKM-MTFD as follows:
 has_Attr_Phe represents the attributes of FPs. The sub-object
property of has_Attr_Phe is also defined to represent a specific
attribute of FP, such as has_Char_Phe, has_Effect_Phe, has_De-

Table 2
Main classes of the core ontology of OKM-MTFD.

Main Class Subclass Note

ZFaultPhenomenon ZFaultPheIntrinsic The intrinsic abstract subclasses of ZFaultPhenomenon
ZFaultPheFunctional The functional abstract subclasses of ZFaultPhenomenon
ZFaultPhenomenonAttribute ZFaultPheAttrCharacteristic The characteristic attribute of FP
ZFaultPheAttrDegree The degree attribute of FP
ZFaultPheAttrFreq The frequency attribute of FP
ZFaultPheAttrAlarmType The alarm type attribute of FP
...
ZFaultMaintenance ZFaultMntIntrinsic The intrinsic abstract subclasses of ZFaultMaintenance
ZFaultMntFunctional The functional abstract subclasses of ZFaultMaintenance
ZFaultMaintenanceAttribute ZFaultMntAttrCharacteristic The characteristic attribute of FM
ZFaultMntAttrType The type attribute of FM
ZFaultMntAttrEmergencyDegree The emergency degree attribute of FM
ZFaultMntAttrLevel The level attribute of FM
ZFaultMntAttrDuration The duration attribute of FM
...
ZFaultCause ZFaultCauElectric/ZFaultCauEnv/. . . The subclasses of ZFaultCause
ZFaultCauseAttribute ZFaultCauAttrFrequency The frequency attribute of FC
ZFaultCauAttrDegree The degree attribute of FC
...
ZFaultLocation – The subclasses of ZFaultLocation
ZFaultAuxiliaryAttribute ZFaultAuxAttrCNC The CNC related attribute of auxiliary fault attribute
ZFaultAuxAttrLing The linguistics related attribute of auxiliary fault attribute
...
100 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112

type
degree

attribute
location

has_Attr_Cau

Fault Cause

Fault Fault
Phenomenon has_Mnt_Phe Maintenance
has_Attr_Phe has_Attr_Mnt

location attribute attribute location

mapping
has_Phe_Loc has_Location_Mnt

effect characteristic characteristic level

Fig. 6. Relationships of the core ontology of OKM-MTFD.

gree_Phe, has_DetMethod_Phe, etc. Particularly, MainPheChar_is,
which is a sub-object property of has_Attr_Phe, is defined to
relate the intrinsic FPA and each FPA must have only one Main-
PheChar_is object property. Similarly, has_Attr_Mnt is defined to
represent the attributes of FM.
 has_Location represents the FL. As the locations where FP, FM
and FC happened may vary for even one record, the sub-
object properties has_Location_Mnt, has_Location_Phe, has_Loca-
tion_Cau are defined and the inverse object property has_Loca-
tion_inv is defined.
 has_Cause represents the FC. Practically, the sub-object proper-
ties has_Cause_Phe and has_Cause_Mnt are preferred to relate FP
and FM, as they are more targeted. Moreover, the inverse object
property has_Cause_inv is also defined for reverse reasoning or
querying.
 The FP and FM are related by has_Mnt_Phe and its inverse object
property has_Phe_Mnt, which are important object properties
for reasoning and querying from FP to FM or vice versa.
Table 3 lists the main object properties of the core ontology of
OKM-MTFD, the sub-object properties of which can be defined,
adapted, extended and modified in practical use. The data property
is used to relate the individuals with data, which are often needed
for specific individuals, so it will be not discussed further within
this article.
4.2.3. Hierarchy of the core ontology of OKM-MTFD
The relations between classes or between class and attribute are
defined using axioms provided in OWL and the core ontology of
OKM-MTFD is established as shown in Fig. 7, which is plotted using
the OntoGraf plugin of Protégé. The comprehensive usage of Sub-
ClassOf, EquivalentTo and DisjointWith axioms are beneficial to
build a sound, robust and semantic-rich OKM-MTFD and knowl-
edge base with the ability of classification, inheritance and
reasoning.
Axiom SubClassOf can be used to define the inheritance relation-
ship between classes, for auxiliary classification and to express the
attributes of the class and the relationships between classes. This
paper adopts axiom SubClassOf to establish the hierarchy structure
of the classes in OKM-MTFD. For defining attributes of the fault ele-
ment, an anonymous class is built first through the verb-object
construction association of the object property and the class of
fault element attribute. Then, the class of fault element inherits
the anonymous class using the axiom SubClassOf and thus obtains
the attribute and the semantics, which can be identified, under-
stood and reasoned by a reasoner. Moreover, some, only and exactly
can be used to define restrictions for object property or data prop-
erty, ensuring that the values of all the individuals belonging to the
fault element class are individuals of the specified fault element
attribute class in the anonymous class within the restrictions.
The relationships between classes can be defined in the same way.
Axiom EquivalentTo can be used to define the equivalence rela-
tionship between classes and as necessary and sufficient condi-
tions for reasoning or reverse derivation, which is helpful for
finding other equivalent properties. In this paper, it is mainly used
for re-classification in the two-stage classification method of fault
elements and referring to ontology in the external ontology refer-
ence method. Axiom EquivalentTo can be used to distinguish the

Table 3
Main object properties of the core ontology of OKM-MTFD.

Object property Domain Range Note

has_Attr_Phe ZFaultPhenomenon ZFaultPhenomenonAttribute Attribute of FP
has_Attr_Mnt ZFaultMaintenance ZFaultMaintenanceAttribute Attribute of FM
has_Attr_Cau ZFaultCause ZFaultCauseAttribute Attribute of FC
has_Attr_Aux – ZFaultAuxiliaryAttribute Auxiliary attribute for fault element
has_Cause ZFaultPhenomenon, ZFaultMaintenance ZFaultCause The cause of FP and FM
has_Cause_inv ZFaultCause ZFaultPhenomenon, ZFaultMaintenance The possible FP or FM may relate to the FC
has_Location ZFaultPhenomenon, ZFaultMaintenance ZFaultLocation The FL of FP, FM and FC
ZFaultCause
has_Location_inv ZFaultLocation ZFaultPhenomenon, ZFaultMaintenance The possible FP, FM or FC may happen at the
ZFaultCause FL
has_Mnt_Phe ZFaultPhenomenon ZFaultMaintenance The FM method for FP
has_Phe _Mnt ZFaultMaintenance ZFaultPhenomenon The FP may be dealt with by the FM
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
Fig. 7. Core ontology of OKM-MTFD.
concepts classes and act as restrictions. In this paper, it is mainly
used to precisely define the fault element or attribute classes,
ensuring the consistency of OKM-MTFD.
For example, the intrinsic class of FP ZFaultPheIntr_Vibration has
attributes such as FL, FC, and FM and is independent of other
intrinsic classes. By using the axioms SubClassOf and DisjointWith
and object properties has_Cause_Phe, has_Location_Phe and
has_Mnt_Phe defined above, the predicate calculus formula is as
follows:
SubClassOf:
ZFaultPheIntrinsic
has_Location_Phe some (ZSubSysBody or
ZSubSysTransmission)
has_Cause_Phe some ZFaultCauMechanic
has_Mnt_Phe some ZFaultMntCau_Mechanic
DisjointWith:
ZFaultPheIntr_Noise, ZFaultPheIntr_Vibration,
ZFaultPheIntr_Damage, . . .
4.3. Two-stage classification method of fault elements
Due to various fault elements such as FP and FM, one single
classification method has difficulty describing its characteristics,
so this paper proposes a two-stage classification method of fault
elements to meet different needs and research perspectives. The
method classifies FP elements and FM elements in two stages. A
fault element will be initially classified by intrinsic attribute, i.e.
intrinsic class; then, a fault element will have a secondary stage
of classification by different configurations of fault attributes, i.e.
functional classes. The two-stage classification method of fault ele-
ments provides a hierarchical knowledge representation of fault
elements, which improves the customization, flexibility and exten-
sibility of KMM-MTFD.
101
For a fault phenomenon, the definitions are as follows:
Def 6. Intrinsic class of a fault phenomenon: An abstract class can
only represent the intrinsic quality of a fault phenomenon, which
should be used for initial classification. It only has one fault
phenomenon intrinsic attribute and may have multiple fault
phenomenon characteristics or general attributes. Each fault
phenomenon can only belong to one intrinsic class of fault
phenomenon.
Def 7. Functional class of a fault phenomenon: A logic class that
has multiple fault phenomenon attributes, which should be used
for a particular purpose or function. It is a re-classification for an
intrinsic class of fault phenomenon and will be accurately defined
in detail by different combinations of fault phenomenon attributes
using predicate logic. Each fault phenomenon can belong to multi-
ple functional classes of fault phenomenon.
In the ontology model, the intrinsic class can be defined as hav-
ing only one intrinsic attribute using axiom EquivalentTo. In practi-
cal applications, the intrinsic class of FP can be obtained in Table 4,
or by referring to the customized intrinsic classes for a specific
machine tool. Generally, an FP can be classified into an existing
intrinsic class of FP according to its intrinsic attribute. However,
pre-defined intrinsic classes may not contain all FPs, and thus
OKM-MTFD must provide support for defining new intrinsic
classes of FP.
To satisfy various needs and research perspectives, it is neces-
sary to define appropriate functional classes. The functional class
can be defined equivalently to a set or sets of attribute classes
using axiom EquivalentTo. In ontology, SubClassOf (CE1 CE2) means
that CE1 is a subclass of CE2 and that CE1 will inherit any axioms or
relations belonging to CE2. Therefore, intrinsic classes or individu-
als of FP can be re-classified into the corresponding functional
classes according to its fault element attributes and thus inherit
the functional classes’ axioms, object properties and data proper-
102 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
Table 4
Intrinsic and functional classes of FP and FM.
Class Subclasses
ZFaultPheIntrinsic ZFaultPheIntr_Scratch
ZFaultPheIntr_Smoke
ZFaultPheIntr_Noise
ZFaultPheIntr_Vibration
ZFaultPheIntr_*
ZFaultPheFunctional ZFaultPheCause
ZFaultPheDegree
ZFaultPheFrequency
ZFaultPheEffect
...
ZFaultMntIntrinsic ZFaultMntIntr_Replace
ZFaultMntIntr_Adjust
ZFaultMntIntr_Repair
ZFaultMntIntr_Restart
ZFaultMntIntr_*
ZFaultMntFunctional ZFaultMntCause
ZFaultMntType
...
ties. As mentioned above, an intrinsic class or individual of FP may
have diverse research perspectives, and thus the functional classes
should not be completely isolated and thus do not need to set
axiom DisjointWith. Therefore, an intrinsic class or individual can
be re-classified into multiple functional classes as needed, which
enables various needs and purposes to be fulfilled. Moreover, exis-
tential restrictions, universal restrictions and cardinality restric-
tions such as only, not and exactly can be used to restrict the
axioms to ensure the correctness of the two-stage classification.
Fig. 8 illustrates the relations between the intrinsic class and
functional class of FP. For example, FP individual A, can be initially
classified into intrinsic class I within intrinsic attribute AttrIntrI, and
it has the characteristic attribute AttrCI and general attributes AttrI
and AttrII. Functional class I is defined as having FPA AttrCI; func-
tional class II is defined as having FPA AttrIntrI, AttrCI and AttrI; func-
tional class III is defined as having FPA AttrI and AttrII; functional
class IV is defined as having FPA AttrIntrII and AttrII. After reasoning,
intrinsic class I will be re-classified into functional class I and II, thus
inherit relationship R1, R2 and R3. FP individual A will be re-
classified into functional class I, II and III and thus inherit relation-
ship R1, R2, R3 and R4 and it will not be a subclass to functional class
IV without FPA AttrIntrII. At the same time, FP individual B can be
initially classified into intrinsic class I and re-classified later into
functional class I. FP individual C has FPA AttrIntrII and will be re-
classified into functional class IV and thus inherit relationship R5.
Similarly, FM also has intrinsic classes and functional classes of
FM. The intrinsic classes of FM can be defined according to the FM
intrinsic attributes such as repairing, modification and replace-
ment. The functional classes of FM can be defined based on the dif-
ferent combinations of FMA such as type, severity, emergency
degree, etc. This paper has provided some of the intrinsic classes
and functional classes of FP and FM that are the core fault elements
in fault diagnosis problems as shown in Table 4. In practical appli-
cations, the intrinsic classes and functional classes should be estab-
lished according to the attributes and demands of fault diagnosis
for specific machine tools.
4.4. External ontology reference method
To assure the commonality and portability of OKM-MTFD, the
classes such as fault location, fault cause, fault detection method,
intrinsic classes of FP and FM, etc. can be set as interfaces for refer-
ring to the corresponding external ontology, to be more adaptive
for specific machine tools and improve the modularity for con-
Description
Intrinsic FP class for scratch of parts
Intrinsic FP class for sudden smoke of CNC or parts
Intrinsic FP class for sudden noise of CNC or parts
Intrinsic FP class for vibration of CNC or parts
Other possible intrinsic FP class
Functional FP class classified by various FCs, such as mechanical, electrical, etc.
Functional FP class classified by FP degree, which is determined by users
Functional FP class classified by FP frequency, which is determined by users
Functional FP class classified by FP effect, such as part, subsystem, or system level effect
Other user defined functional FP class
Intrinsic FM class for replace parts or assembly
Intrinsic FM class for adjust parts or assembly
Intrinsic FM class for repair parts or assembly
Intrinsic FM class for reboot the CNC
Other possible intrinsic FM class
Functional FM class classified by various FCs, such as mechanic, electric, etc.
Functional FM class classified by FM type, such as minor, general, heavy, etc.
Other user defined functional FM class
structing knowledge models. By using the axiom EquivalentTo,
the interface class and the target class of external ontology should
be equivalent and thus can be inferred by a reasoner. A sample is
shown in Fig. 9.
The proposed external ontology reference method can:
 Improve the reusability of existing knowledge.
 Develop the fault elements such as intrinsic classes of FP and
FM externally for specific machine tools within the common
core ontology proposed above to ensure both integration ability
and scalability.
 Improve the modularity of OKM-MTFD to reduce the complex-
ity of development. In addition, the ontology development can
be implemented and maintained independently for different
machine tools.
For example, the FL class ZFaultLocation can be defined equiva-
lently to the structure ontology of the machine tool ZMTStructure
externally using the axiom EquivalentTo, thus importing the exist-
ing ontology.
Equivalent To:
ZMTStructure
4.5. Construction approach of OKM-MTFD
The core ontology defines the key concepts, properties, relations
and their hierarchy in the fault diagnosis domain; however, more
detailed fault elements should be defined to establish a sound
OKM-MTFD for a type of or a specific machine tool according to
the characteristics. Thus, a construction approach of OKM-MTFD
is proposed based on the classical ontology modelling method.
By virtue of the two-stage classification method and external
ontology reference method, the core ontology of OKM-MTFD will
be extended with good transplantation ability. The approach is
shown in Fig. 10.
First, the predicate-logic-based analysis method of fault ele-
ments proposed in Section 4.1 is adopted to analyse fault elements
and fault element attributes from original fault diagnosis records
for a general type of or a specific machine tool. Thus, the intrinsic
classes of FP and FM should be defined and inherit the ULCs of the
core ontology of OKM-MTFD using axiom SubClassOf, namely,
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112 103

R3 R4
R2

R5
R1 Functional
Functional class II Functional
class I class III

Functional
AttrI class IV
Intrinsic classI
AttrCI AttrII

AttrItri
I AttrItri
II

fault phenomenon fault phenomenon

individual B
fault phenomenon
individual C
individual A

fault phenomenon Attr intrinsic attribute Attr

Instance B general attribute Asserted
individual

fault phenomenon characteristic

Class I Attr R relationship Inferred
class attribute

Fig. 8. Relationship of the intrinsic class and functional class of FP.

External ontology: Auxiliary Attribute

ZFaultAuxiliaryAttribute
ZFaultAuxAttr
CNC

ZFaultAuxAttr
Lang ...

External ontology: Intrinsic Class External ontology: Structure

ZFaultPheIntrinsic
ZMTStructure
ZFaultPheIntr ZFaultPheIntr ZSubSysTrans
_Scratch _Noise mission
ZSubSysBody
ZFaultPheIntr
_Smoke ZSubSysHydraulic
... ...

Fig. 9. Sample of the external ontology reference method.

ZFaultPheIntrinsic and ZFaultMntIntrinsic. Other fault element ontology can be constructed externally by structure experts and
classes, fault element attributes classes and properties should be connected to OKM-MTFD using axiom EquivalentTo. Third, accord-
set as subclasses of the corresponding ULCs. Second, the structure ing to the actual requirements of fault diagnosis, functional classes
104 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112

Core ontology
Of
OKM-MTFD

FP intrinsic class
A type of or a FM intrinsic class OKM-MTFD
specific Machine tool Attribute class With
Axiom New fault elements
...

Structure ontology OKM-MTFD

Of With
External structure ontology
Machine tool

Analysis of FP functional class

Fault diagnosis FM functional class OKM-MTFD
With
SWRL rules
Functional classes and SWRL Rules
...

FP/FM/FC individuals
Analysis of FL individuals OKM-MTFD
Fault element individuals Object properties For
Data properties A type or a specific machine tool
...

Knowledge base
Based on OKM-MTFD For
A type or a specific machine tool

Fig. 10. Construction approach of OKM-MTFD.

of FP and FM can be determined by different combinations of fault
element attributes. Moreover, SWRL rules can be defined according
to the characteristics and the requirements of the diagnosis of
machine tools. To be validated, the consistency of ontology should
be checked and modified if anything goes wrong.
Finally, the core ontology is extended for building OKM-MTFD
and transplanted for a general type or a specific machine tool.
The fault element individuals can then be added to build a knowl-
edge base based on OKM-MTFD. However, the knowledge base
may not contain all fault records or fault element classes of
OKM-MTFD, or they may not be sufficient for classification. More-
over, some new faults would occur due to the complexity of
machine tools and the manufacturing workshops. Therefore, the
maintenance approach of an ontology-based knowledge model
and knowledge base for a specific machine tool are proposed to
add new fault knowledge and update the knowledge model
dynamically. The approach will be discussed based on FP and the
FM and FC can be treated in the same way.
 Acquiring newly added FP knowledge
New FP knowledge, or the undefined knowledge in the knowl-
edge base, will be analysed first using the predicate-logic-based
analysis method of fault elements proposed in Section 4.1. Then
it will be initially classified as an individual of an intrinsic class
according to its intrinsic attribute and then the other attributes
should be added. After reasoning, the newly added individuals will
be re-classified into the pre-defined functional class. Thus, it inher-
its the has_Mnt_Phe attribute and other attributes of the functional
classes.
 Updating the intrinsic class of FP dynamically
For a new FP, which cannot be classified into the pre-defined
intrinsic classes in the OKM-MTFD, a new intrinsic class can be
built as a subclass of ZFaultPheIntrinsic according to its intrinsic
attribute. Then the FP and others that have the same intrinsic attri-
bute that may occur in the future can be individuals of the newly
defined class. If the FPA is also new to OKM-MTFD, the attribute
class should be defined as a subclass of ZFaultPhenomenonAttribute.
Then newly added classes should be updated regularly or irregu-
larly to improve the knowledge model.
 Updating the functional class of FP dynamically
If the functional classes of the FP cannot satisfy the description
or reasoning needs, a new functional class can be defined as a sub-
class of ZFaultPheFunctional using the appropriate attributes group.
As the addition action is performed on to the ontology model and
some existing functional classes may go out-of-date, the mainte-
nance work must be done to ensure the correctness and validity.
In addition, the newly added classes can extend the core ontology
of OKM-MTFD in return.
4.6. Knowledge reasoning and querying methods
The analysis and solving of fault are the core problems of fault
diagnosis. The reasoning and retrieval layer of KMM-MTFD is
designed to address it based on OKM-MTFD as established in this
paper. From the view of the system, a fault diagnosis system
should analyse the FP (input) and provide the results and mainte-
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
nance method (output). Thus, functions such as querying, reason-
ing and guidance for maintenance or special faults should be sup-
ported. The knowledge reasoning methods based on OWL axioms,
SWRL rules and special fault attributes are proposed for reasoning
from multiple perspectives.
 A reasoning method based on OWL axioms
This method mainly uses the re-classification for fault element
classes through the attributes matching of intrinsic classes and
functional classes and thus realizes the analysis and reasoning of
fault elements in the manner of inheriting the functional classes’
attributes. The method depends on the OWL axioms and the
two-stage classification method. It is applicable for consistency
verification and initial analysis for newly added fault knowledge.
For example, an FP individual FaultIndividualA with attributes such
as location, cause and alarm type will be re-classified into func-
tional class FunctionalClassA. Thus, it is initially analysed by inher-
iting the attributes and relationships of FunctionalClassA.
 A reasoning method based on SWRL rules
OWL can only represent property restrictions of one single class,
while the rules in the fault diagnosis domain always contain many
classes, which is beyond the ability of OWL. Moreover, the OWL DL
is based on Description Language and cannot express rules format-
ted in ‘‘if. . .then. . .”. SWRL, which is based on the Horn Clause, can
be used effectively to compensate these deficiencies. SWRL is
ontology-based and can be conveniently applied in an ontology-
based knowledge base using the existing individuals and
properties.
This paper adopts SWRL rules for reasoning and restriction. The
possible FM methods for FP can be reasoned through FCs. Com-
pared to the OWL axiom based reasoning method, it is more flexi-
ble and commonly used for various fault diagnosis needs. The
implicit fault knowledge between multiple individuals should be
discovered using the SWRL rules. Based on the predicate logic,
the fault elements and their individual terms or predicates can
be manipulated to write a SWRL rule.
 A reasoning method based on special fault attributes
This method can be used for the treatment of special fault prob-
lems. For a special fault, the reasoning is done directly to find the
corresponding FM methods contained in intrinsic or functional
classes through the special attributes. For example, if the emer-
gency degree of an FP is very high, the matched FM methods will
be directly obtained such as ‘‘directly notify the expert mainte-
nance personnel”.
 Knowledge querying based on SPARQL
Knowledge querying is a common interactive activity in
knowledge-based system. It is used for querying the basic informa-
tion of fault elements and the results of reasoning, including the
existing and implicit fault knowledge. By using the reasoning
methods above, the implicit fault knowledge should be discovered.
OKM-MTFD has established the relationships and inverse relation-
ships between fault elements by using axioms and object proper-
ties, ensuring bidirectional querying. A querying method based
on SPARQL is proposed for four types of knowledge query prob-
lems, i.e. querying FM from FP, querying FP and FM from FC, query-
ing FP, FM and FC from FL and querying FP from FM. The definition
of query clause can be found in the SPARQL specification [50] pro-
posed by W3C.
105
5. Case study
A case study for KMM-MTFD is carried out on an YS3120CNC5
hobbing machine tool, which is produced by a machine tool man-
ufacturer in Chongqing, China. This paper adopts Protégé 4.3 as the
ontology-modelling tool. With the core ontology of OKM-MTFD
and the construction approach, an ontology-based knowledge
model and knowledge base of the hobbing machine tool are estab-
lished. Moreover, knowledge reasoning and querying are demon-
strated to show the validity of the proposed KMM-MTFD.
5.1. Construction of the knowledge model and knowledge base for
YS3120CNC5
The original data of YS3120CNC5 such as the structure, specifi-
cation and historical experience and records of fault diagnosis are
obtained after full investigation and research. According to the
construction approach of OKM-MTFD proposed in Section 4.5, the
original fault diagnosis records are analysed using the predicate-
logic-based analysis method of fault elements proposed in Sec-
tion 4.1. Due to the length of the paper, we take some typical faults
that occur at ‘‘big column” as examples to present the KMM-MTFD
for YS3120CNC5.
The original fault diagnosis records from the machine tool man-
ufacturer are analysed using the predicate-logic-based analysis
method. For example, for the FP ‘‘scratch of the z axis linear guide”,
the properties formatted like ‘‘Subject-Verb-Object” are extracted
as ‘‘intrinsic characteristic attribute is scratch”, ‘‘(the FP) occur at
z axis linear guide”, ‘‘severity FPA is slight”, ‘‘frequency FPA is gen-
eral”, ‘‘alarm type is ‘none’” and ‘‘detection method is ‘see’”. The
other fault elements can be analysed in the same way and then
semi-structured knowledge of fault elements is acquired as
showed in Tables 5–8.
From the above knowledge tables, the classes, properties and
relationships of fault elements are obtained as follows:
 For FP, the characteristic attributes listed above are their
intrinsic attributes. The intrinsic classes ZFaultPheIntr_Scratch,
ZFaultPheIntr_Dither, ZFaultPheIntr_Slip, and ZFaultPheIntr_Shift
are established using object property MainPheChar_is. Other
attribute classes such as characteristics ZFaultPheAttrCharacter-
istic, severity ZFaultPheAttrSeverity, frequency ZFaultPheAttrFreq,
alarm type ZFaultPheAttrAlarmType, detection method
ZFaultPheAttrDetectMethod and degree ZFaultPheAttrDegree are
established as subclasses of ZFaultPhenomenonAttribute.
 For FM, the intrinsic classes ZFaultMntIntr_Repair,
ZFaultMntIntr_Replace, and ZFaultMntIntr_Adjust and the
attribute classes such as characteristics ZFaultMntAttrCharacter-
istic, emergency degree ZFaultMntAttrEmergencyDegree, level
ZFaultMntAttrLevel and type ZFaultMntAttrType, are established
as subclasses of ZFaultMaintenanceAttribute.
 For FC, the classes ZFaultCau_Damage and ZFaultCau_Loose, are
established as subclasses of ZFaultPheCau_Mechanic. The attri-
bute classes such as characteristics ZFaultMntAttrCharacteristic,
degree ZFaultCauAttrDegree and frequency ZFaultCauAttrFre-
quency are established as subclasses of ZFaultCauseAttribute.
 For FL, the paper constructs the ontology of the structure of
YS3120CNC5 in the hierarchy of ‘‘subsystem – assembly –part”
externally. The top class ZMTStructure is related to the ULC
ZFaultLocation using axiom EquivalentTo. The main classes of
the ontology of the structure of the YS3120CNC5 hobbing
machine tool are shown in Fig. 11.
This case adopts the top class ZMTFault and extends the core
ontology of OKM-MTFD with the established fault element classes
106 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112

Table 5
Knowledge table of FP of the hobbing machine tool (Partial).

No. Fault phenomenon Fault phenomenon attribute Auxiliary

attribute
Description Char Location Severity Freq Alarm Detection Degree ...
FL205FP0001 Scratch of the z axis linear guide Scratch z axis linear guide Slight General None See Obvious ...
FL205FP0002 Dither of the slide plate Dither slide plate General General None Observe Obvious ...
FL205FP0003 Scratch of the x axis linear guide Scratch x axis linear guide Slight General None See Obvious ...
FL205FP0004 Dither of the big column Dither Big column Serious Sudden None See Obvious ...
FL205FP0005 Slip of the gear belt transmission assembly Slip Gear belt Slight General None See General ...
FL205FP0006 Shift of the expansion sleeve Shift Expansion sleeve General Sudden None Observe General ...

Table 6
Knowledge table of FM of the hobbing machine tool (Partial).

No. Fault maintenance Fault maintenance attribute Auxiliary attribute

Description Char Location Emergency Level Type ...
FL205FM0001 Repair the z axis linear guide Repair z axis linear guide Emergent Note and report Occurrence ...
FL205FM0002 Replace the z axis linear guide Replace z axis linear guide Extremely emergent Warning and report Occurrence ...
FL205FM0003 Adjust the tuner screw Adjust Tuner screw General Note and process Occurrence ...

Table 7
Knowledge table of FC the hobbing machine tool (Partially).

No. Fault cause Fault cause attribute Auxiliary attribute

Description Char Location Degree Frequency ...
FL205FC0001 Damage of the z axis linear guide Damage z axis linear guide Serious General ...
FL205FC0002 Loose of the z axis linear guide Loose z axis linear guide General Frequent ...
FL205FC0003 Loose of the tuner screw Loose Tuner screw General General ...

Table 8 5.2.1. Reasoning based on OWL axioms

Knowledge table of FL of the hobbing machine tool (Partial). The functional class is defined to describe the fault element
No. Fault location Assembly Subsystem from multiple-perspectives and can be used for re-classification
FL205S001 z axis linear Big column Body using a reasoner. For example, a functional class FP ZFaultPheCau_
guide Mechanic is defined as ‘‘having mechanic cause”, ‘‘occurred at the
FL205S002 Slide plate Big column Body body and transmission subsystems” and ‘‘having a mechanic type fault
FL205S003 Tuner screw Gear belt transmission Transmission maintenance method”. The expression is as follows:
assembly

EquivalentTo:
has_Cause_Phe some ZFaultCauMechanic
has_Location_Phe some (ZSubSysBody or
and fault element attribute classes. Moreover, some functional ZSubSysTransmission)
classes for FP and FM are defined for demonstration. Thus, OKM- SubClassOf:
MTFD for the hobbing machine tool YS3120CNC5 is established has_Mnt_Phe some ZFaultMntCau_Mechanic
as shown in Fig. 12.
The fault diagnosis knowledge of YS3120CNC5 listed above can
be added to OKM-MTFD under the Individual Tab in Protégé 4.3.
In the Description View, individuals can be asserted as instances The intrinsic class ZFaultPheIntr_Scratch is defined as:
of a class through the Types label. The independence of individu-
als can be asserted through the Different Individuals label. In the EquivalentTo:
Property Assertions View, the relationships of individuals and ZFaultPheIntrinsic and (MainPheChar_is only
their attributes can be asserted through the Object Property ZFaultPheAttrChar_Scratch)
Assertions label. Once the individuals and their relationships are SubClassOf:
added and defined, the knowledge base is constructed as shown ZFaultPheIntrinsic
in Fig. 13. has_Cause_Phe some ZFaultCauMechanic
has_Degree_Phe some ZFaultPheAttrDeg_Slight

5.2. Reasoning in the knowledge base of YS3120CNC5

The reasoning demonstration is carried out using the HermiT
1.38 reasoner to show the effect of solving the reasoning problem
of fault diagnosis with the proposed KMM-MTFD and the estab-
lished OKM-MTFD and knowledge base. The reasoning process dis-
covers the implicit semantics of the knowledge base and thus
improve the value of knowledge.
After reasoning, the intrinsic class ZFaultPheIntr_Scratch is classified
into the functional classes ZFaultPheCau_Mechanic and ZFaultPhe-
Deg_Slightly, as shown with label ‘‘1” in Fig. 14. Thus, class ZFault-
PheIntr_Scratch inherits the attribute ‘‘has_Mnt_Phe some
ZFaultMntCau_Mechanic” of class ZFaultPheCau_Mechanic as shown
with label ‘‘2”, which means, ‘‘having a mechanic type fault mainte-
nance method”. In the same manner, the other intrinsic classes such
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112 107

Fig. 11. Ontology model of the structure of YS3120CNC5.

Overview Fault Phenomenon Fault Maintenance Object Property

Fig. 12. OKM-MTFD for YS3120CNC5.

as ZFaultPheIntr_Damage, ZFaultPheIntr_Dither and ZFaultPheIntr_ 5.2.2. Reasoning based on SWRL rules

Leakage will be re-classified into the corresponding functional
classes as shown via label ‘‘3”. Moreover, the individuals will also
be classified into the functional classes according to their attributes
and thus be initially analysed, as shown with label ‘‘4” and label ‘‘5”.
Note that, the items with shaded background are inferred and the
others with blank background are asserted.
In this case, a ‘‘one-to-one” diagnosis problem is studied using
the reasoning method based on SWRL rules. For example, given
that the intrinsic class of FP ZFaultPheIntrinsic and the intrinsic
class of FM ZFaultMntIntrinsic have the same FC ZFaultCause, then
ZFaultPheIntrinsic and ZFaultMntIntrinsic have object property
has_Mnt_Phe. The expression is:
108 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112

Fault elements Fault element attributes Auxiliary attributes Struture of YS3120CNC5

Fig. 13. Knowledge base of fault diagnosis for YS3120CNC5.

Fig. 14. Reasoning results based on OWL axioms.

 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112 109

?x ?has_Mnt_Phe ?y.
SWRL Rule:
?has_Mnt_Phe rdfs:label ?Labelhas_Mnt_Phe.
ZFaultMntIntrinsic(?fm), ZFaultPheIntrinsic(?fp),
?has_Mnt_Phe zmt:comment_eng ?
ZFaultCause(?fcm), ZFaultCause(?fcp),
Comment_enghas_Mnt_Phe.
has_Cause_Phe(?fp, ?fcp), has_Cause_Mnt(?fm, ?
?x rdfs:label ?Labelx.
fcm), SameAs (?fcp, ?fcm) -> has_Mnt_Phe(?fp, ?fm)
?x zmt:comment_eng ?Comment_engx.
?y rdfs:label ?Labely.
?y zmt:comment_eng ?Comment_engy.
After reasoning, the FP is related to those with the same FC through
}
object property has_Mnt_Phe. For example, the FP ‘‘Scratch of the Z
axis linear guide” has FM ‘‘Repair the Z axis linear guide” after reason-
ing and thus fault diagnosis is accomplished as shown in Fig. 15. The
The result in Fig. 16 shows that the FP has the FM ‘‘Repair the X axis
equivalent and super object properties are reasoned out as well. In
linear guide”. In the same manner, the query problems such as
addition, Fig. 15 also shows that names for a class or individual can
querying FM from FP, querying FP and FM from FC, querying FP,
be customized without changing its semantic.
FM and FC from FL and querying FP from FM, can be queried using
SPARQL. The results of querying for the FC ‘‘Loose of the tuner screw”,
5.3. Querying in the knowledge base of YS3120CNC5
the FL ‘‘X axis linear guide” and the FM ‘‘Repair the Z axis linear guide”
are shown in Figs. 17–19, respectively for demonstration.
The querying demonstration is carried out using Apache Jena
Fuseki [52]. Apache Jena Fuseki is an SPARQL server. It can run as
6. Comparison and discussion
an operating system service, as a Java web application (WAR file),
and as a standalone server. In this paper, OWL/RDF is applied to
This paper adopts a semantic-network-based knowledge repre-
represent the knowledge such that inter-exchange, retrieval and
sentation method and defines the key concepts of fault elements
storage become possible among fault diagnosis applications in
multidisciplinary environment. After essential configurations, the independently. Thus, fault knowledge can be defined using OWL
axioms to express the relations according to practical needs,
established knowledge base can be uploaded to the Dataset of
Fuseki in RDF/XML format. It will be successful if the RDF/XML file improving the flexibility and reusability of knowledge. Compar-
isons among similar studies from a knowledge engineering per-
is validated. For the FP ‘‘Scratch of the X axis linear guide”, the label
of which is ‘‘FL205FP0003”, the SPARQL query clause is: spective are listed in detail:

(1) Regarding knowledge representation: Paper [12] proposes a

SPARQL Query:
method of intelligent fault diagnosis based on ontology for
SELECT DISTINCT ?Labelx ?Comment_engx ?
wind turbines and adopt a framework representation
Comment_enghas_Mnt_Phe ?Comment_engy
method for constructing ontologies. The framework-like
WHERE {
representation method could directly define the relations
?x rdfs:label ‘‘FL205FP0003”.
between key concepts, but the flexibility and extensibility
?has_Mnt_Phe rdfs:label ‘‘has_Mnt_Phe”.
decreased and some situations might not need all the infor-
mation. For example, class ‘‘Failuremode” and its subclasses

Fig. 15. Reasoning results based on SWRL rules.

110 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112

Fig. 16. Result of querying FM from FP.

Fig. 17. Result of querying FP and FM from FC.

Fig. 18. Result of querying FP, FM and FC from FL.

Fig. 19. Result of querying FP from FM.

are strongly interrelated, which means that once the inner
element changes, the original fault mode instance and its
relations with other instances may no longer be valid. In this
paper, the key concepts are treated as the nodes of the
semantic network. Thus, the semantics remain the same
even when some attributes are modified and the new prob-
able hidden knowledge can be inferred under new situa-
tions, improving adaptability and reusability.
(2) Regarding knowledge integration: Paper [10] proposes a
fleet-wide knowledge model for an entire fleet system in
the marine domain and context ontologies of general con-
cepts for related domain are established, improving the
knowledge extensibility and integration capability. This
paper concentrates on the relations of fault elements and
cares less about related domains such as service mode, oper-
ational environment, application, etc. However, these
related ontologies can be linked with OKM-MTFD using the
external ontology reference method in the future.
(3) Regarding knowledge reasoning: The established ontology
models in many literatures [9,10,12,34,39] support SWRL
rules, and so does OKM-MTFD. Moreover, the proposed rea-
soning method based on OWL axioms can check the consis-
tency of fault knowledge and analyse the newly added FPs
preliminarily and the one based on special fault attributes
supports fast finding of specific faults, providing flexible,
customizable and powerful methods to utilize knowledge
efficiently.
(4) Regarding implementation of knowledge model: Paper
[12] established an ontology model for wind turbines,
which can be implemented for similar wind turbines.
However, if there are any new requirements or transplan-
tation for other machineries, the model might need a
major revision. This paper has constructed the core ontol-
ogy model for fault diagnosis and provided the
extension method for specific machine tools or other
machineries.
 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
It is worth mentioning that the method proposed in this paper
requires massive preliminary knowledge to establish the semantic
network for a new machine. The construction process in the initial
phase might be challenging and require immense human efforts.
Thus, the examples in Section 5 are relatively simple only to show
the characteristics of KMM-MTFD. More fault knowledge will be
added later to show the advantages.
7. Conclusions and future work
A knowledge modelling methodology for fault diagnosis of
machine tools based on formal semantics (KMM-MTFD) is pro-
posed in this paper, which is a systemic method for rapid cus-
tomization and integration of OKM-MTFD. The established core
ontology provides a general semantic model in the fault diagnosis
domain of machine tools and a construction approach of OKM-
MTFD is proposed to establish a sound OKM-MTFD for a type of
or a specific machine tool according to the characteristics. Thus,
the core ontology of OKM-MTFD will be extended with good trans-
plantation capability.
Moreover, several methods are proposed for applying and uti-
lizing semantic-network-based knowledge more efficiently:
 The predicate-logic-based analysis method of fault elements
enables the definition of a unified expression for fault elements
with formal semantics.
 The two-stage classification method of fault elements and
external ontology reference method provide hierarchical and
modular knowledge management for various fault phenomena
or fault maintenance, support analysis and diagnosis of newly
added fault phenomena and improve the versatility, practicabil-
ity and extensibility of OKM-MTFD.
 The knowledge reasoning methods based on OWL axioms,
SWRL rules and special fault attributes are available to a rea-
soner to conduct knowledge reasoning of fault diagnosis, anal-
ysis of newly added fault phenomena and treatment of special
faults.
 A knowledge querying method based on SPARQL is proposed for
four types of knowledge querying problems.
The complete work in this paper lays the foundation for an
open, shared, and scalable knowledge model for fault diagnosis
of machine tools that supports establishment of an intelligent fault
diagnosis system. In future works, enhancement of the knowledge
model and other issues relating to intelligent fault diagnosis will be
addressed:
 The construction of a common library for fault diagnosis. The
extraction method for massive fault records will make KMM-
MTFD more practical. The enlargement of common ontology
classes, attributes, rules and instances will support the estab-
lishment of a robust semantic network.
 The knowledge model of fault diagnosis for different levels of a
machine tool based on the core ontology proposed in this paper.
 The intelligent acquisition of fault phenomena. Through the
knowledge representation method used in this paper, a fault
phenomenon is an individual element that can be easily con-
nected to the results of pattern recognition algorithms. The
detection diagnostic method based on signal analysis and intel-
ligent algorithms such as ANN, HMM, SVM can be integrated
with OKM-MTFD.
 Correctness evaluation of the results and performance evalua-
tion of diagnostic procedures should be further studied. The
probability sequence of the inferred results should be
calculated.
111
 The proposed knowledge modelling methodology can also be
applied to other domains of machine tools, such as structure,
cutting tools management, running status monitoring, etc. With
the knowledge model of more machine tool related domains
established, reusability and knowledge volume will be greatly
increased, which will eventually improve the intelligence level.
Acknowledgements
This work is supported by the ‘‘2015 Smart manufacturing pro-
ject of China”. We are also grateful to the anonymous referees and
the editor-in-chief for their suggestions to improve this paper.
References
[1] F. Liu, J. Xie, S. Liu, A method for predicting the energy consumption of the
main driving system of a machine tool in a machining process, J. Clean. Prod.
105 (2015) 171–177.
[2] L. Ramos, Semantic Web for manufacturing, trends and open issues, Comput.
Ind. Eng. 90 (2015) 444–460.
[3] A. Ikram, U. Qamar, Developing an expert system based on association rules
and predicate logic for earthquake prediction, Knowl.-Based Syst. 75 (2015)
87–103.
[4] X. Zhong, H. Fu, Y. Jiang, Coupling ontology with rule-based theorem proving
for knowledge representation and reasoning, Commun. Comput. Inform. Sci.
118 (2010) 110–119.
[5] C. Jun, H. Qinglan, Cost reduction Frame knowledge representation system in
product design process, in: 2010s WRI Global Congress on Intelligent Systems
(GCIS), vol. 2, 2007, pp. 177–180.
[6] P. Witherell, S. Krishnamurty, I.R. Grosse, J.C. Wileden, Improved knowledge
management through first-order logic in engineering design ontologies, Artif.
Intell. Eng. Des. Anal. Manuf. 24 (2010) 245–257.
[7] S. Ma, L. Tian, Ontology-based semantic retrieval for mechanical design
knowledge, Int. J. Comput. Integr. Manuf. 28 (2014) 226–238.
[8] A. Franceschelli, Ontological conceptualization based on the SKOS, J. Comput.
Inform. Sci. Eng. 14 (2014) 672–682.
[9] V. Ebrahimipour, S. Yacout, Ontology-based knowledge platform to support
equipment health in plant operations, in: V. Ebrahimipour et al. (Eds.),
Ontology Modeling in Physical Asset Integrity Management, Springer
International Publishing, Cham, 2015, pp. 221–255.
[10] G. Medina-Oliva, A. Voisin, M. Monnin, J. Leger, Predictive diagnosis based on a
fleet-wide ontology approach, Knowl.-Based Syst. 68 (2014) 40–57.
[11] T. Stålhane, FMEA, HAZID, and ontologies, in: V. Ebrahimipour et al. (Eds.),
Ontology Modeling in Physical Asset Integrity Management, Springer
International Publishing, Cham, 2015, pp. 45–85.
[12] A. Zhou, D. Yu, W. Zhang, A research on intelligent fault diagnosis of wind
turbines based on ontology and FMECA, Adv. Eng. Inform. 29 (2015) 115–125.
[13] B. Shen, S.Y. Zhao, J.H. Wang, Ontology-based fault diagnosis knowledge
representation of CNC machine tool, Appl. Mech. Mater. 427–429 (2013)
1372–1375.
[14] B. Boashash, S. Ouelha, Automatic signal abnormality detection using time-
frequency features and machine learning: a newborn EEG seizure case study,
Knowl.-Based Syst. 106 (2016) 38–50.
[15] C. Wu, T. Chen, R. Jiang, L. Ning, Z. Jiang, A novel approach to wavelet selection
and tree kernel construction for diagnosis of rolling element bearing fault, J.
Intell. Manuf. 1–12 (2015).
[16] Y. Lei, J. Lin, Z. He, M.J. Zuo, A review on empirical mode decomposition in fault
diagnosis of rotating machinery, Mech. Syst. Signal Process. 35 (2013) 108–
126.
[17] W. Chine, A. Mellit, V. Lughi, A. Malek, G. Sulligoi, A. Massi Pavan, A novel fault
diagnosis technique for photovoltaic systems based on artificial neural
networks, Renew. Energy 90 (2016) 501–512.
[18] X. Zhang, B. Wang, X. Chen, Intelligent fault diagnosis of roller bearings with
multivariable ensemble-based incremental support vector machine, Knowl.-
Based Syst. 89 (2015) 56–85.
[19] X. Tang, L. Zhuang, J. Cai, C. Li, Multi-fault classification based on support
vector machine trained by chaos particle swarm optimization, Knowl.-Based
Syst. 23 (2010) 486–490.
[20] Z. Li, W. Ding, A novel fault diagnosis method for gear transmission systems
using combined detection technologies, Res. J. Appl. Sci., Eng. Technol. 6 (2013)
3354–3358.
[21] B. Cai, H. Liu, M. Xie, A real-time fault diagnosis methodology of complex
systems using object-oriented Bayesian networks, Mech. Syst. Signal Process.
80 (2016) 31–44.
[22] Z. Xu, J. Xuan, T. Shi, B. Wu, Y. Hu, A novel fault diagnosis method of bearing
based on improved fuzzy ARTMAP and modified distance discriminant
technique, Expert Syst. Appl. 36 (2009) 11801–11807.
[23] H. Liu, X. Dong, Z. Yang, K. Zheng, The application of intelligent fuzzy inference
to the fault diagnosis in pitch-controlled system, Energy Procedia 16 (2012)
1839–1844.
112 Q. Zhou et al. / Advanced Engineering Informatics 32 (2017) 92–112
[24] H. Wen, Y. Zhen, H. Zhang, A. Chen, D. Liu, An ontology modeling method of
mechanical fault diagnosis system based on RSM, in: Fifth International
Conference on Semantics, Knowledge and Grid, 2009. SKG 2009, 2009.
[25] D. Wang, W.H. Tang, Q.H. Wu, Ontology-based fault diagnosis for power
transformers, in: IEEE PES General Meeting, 2010, pp. 1–8.
[26] C. Nan, F. Khan, M.T. Iqbal, Real-time fault diagnosis using knowledge-based
expert system, Process Saf. Environ. 86 (2008) 55–71.
[27] L. Shu-Hsien, Expert system methodologies and applications—a decade review
from 1995 to 2004, Expert Syst. Appl. 28 (2005) (1995) 93–103.
[28] Y. Wang, Q. Li, M. Chang, H. Chen, G. Zang, Research on fault diagnosis expert
system based on the neural network and the fault tree technology, Procedia
Eng. 31 (2012) 1206–1210.
[29] Y. Zhi-Ling, W. Bin, D. Xing-Hui, L. Hao, Expert system of fault diagnosis for
gear box in wind turbine, Syst. Eng. Procedia 4 (2012) 189–195.
[30] N. Dendani-Hadiby, M.T. Khadir, A case based reasoning system based on
domain ontology for fault diagnosis of steam turbines, Int. J. Hybrid Inform.
Technol. 5 (2012) 89–104.
[31] B. Yang, T. Han, Y. Kim, Integration of ART-Kohonen neural network and case-
based reasoning for intelligent fault diagnosis, Expert Syst. Appl. 26 (2004)
387–395.
[32] E. Olsson, P. Funk, M. Bengtsson, Fault diagnosis of industrial robots using
acoustic signals and case-based reasoning, in: P. Funk et al. (Eds.), Advances in
Case-Based Reasoning, Springer Berlin Heidelberg, Berlin, Heidelberg, 2004,
pp. 686–701.
[33] A.T. Bimba, N. Idris, A. Al-Hunaiyyan, R.B. Mahmud, A. Abdelaziz, S. Khan, V.
Chang, Towards knowledge modeling and manipulation technologies: a
survey, Int. J. Inform. Manage. 36 (2016) 857–871.
[34] V. Ebrahimipour, S. Yacout, Ontology Modeling in Physical Asset Integrity
Management, Springer International Publishing, Cham, 2015.
[35] T.R. Gruber, A translation approach to portable ontology specifications, Knowl.
Acquisit. 5 (1993) 199–220.
[36] L. Liu, W. Yang, Z. Yang, L. Liu, Z. Yang, Ontology model and application
for the CNC integrated detection, Int. J. Comput. Int. Sys. 21 (2007)
378–388.
[37] OWL 2 Web Ontology Language Document Overview, second ed., 2012.
<https://www.w3.org/TR/owl2-overview/> (accessed 2016-08-14).
[38] Ian Horrocks, Peter F. Patel-Schneider, Harold Boley, Said Tabet, Benjamin
Grossof, SWRL: A Semantic Web Rule Language Combining OWL and RuleML,
2004. <https://www.w3.org/Submission/SWRL/> (accessed 2016-02-25).
[39] Y. Zhang, X. Luo, Y. Zhao, H. Zhang, An ontology-based knowledge framework
for engineering material selection, Adv. Eng. Inform. 29 (2015) 985–1000.
[40] N.F. Doherty, A.K. Kochhar, R. Main, Knowledge-based approaches to fault
diagnosis: a practical evaluation of the relative merits of deep and shallow
knowledge, Proc. Inst. Mech. Eng., Part B: J. Eng. Manuf. 208 (1994) 39–45.
[41] J. Lee, F. Wu, W. Zhao, M. Ghaffari, L. Liao, D. Siegel, Prognostics and health
management design for rotary machinery systems—reviews, methodology and
applications, Mech. Syst. Signal Process. 42 (2014) 314–334.
[42] D. Dou, J. Yang, J. Liu, Y. Zhao, A rule-based intelligent method for fault
diagnosis of rotating machinery, Knowl.-Based Syst. 36 (2012) 1–8.
[43] D. Dou, S. Zhou, Comparison of four direct classification methods for intelligent
fault diagnosis of rotating machinery, Appl. Soft Comput. 46 (2016) 459–468.
[44] S. Sahin, M.R. Tolun, R. Hassanpour, Hybrid expert systems: a survey of current
approaches and applications, Expert Syst. Appl. 39 (2012) 4609–4617.
[45] Q. Zhou, K. Liu, Fault diagnosis of CNC machine tools common, Adv. Mater. Res.
317–319 (2011) 1882–1885.
[46] Wikipedia, Failure Mode, Effects, and Criticality Analysis, 2015. <https://en.
wikipedia.org/wiki/Failure_mode,_effects,_and_criticality_analysis#cite_note-
16> (accessed 2016-1-2).
[47] L. Jun, X. Huibin, Reliability analysis of aircraft equipment based on FMECA
method, Phys. Procedia 25 (2012) 1816–1822.
[48] H. Arabian-Hoseynabadi, H. Oraee, P.J. Tavner, Failure modes and effects
analysis (FMEA) for wind turbines, Int. J. Electr. Power Energy Syst. 32 (2010)
817–824.
[49] N. Snooke, C. Price, Automated FMEA based diagnostic symptom generation,
Adv. Eng. Inform. 26 (2012) 870–888.
[50] T.W.C.S. Group, SPARQL 1.1 Overview, 2013. <https://www.w3.org/TR/
sparql11-overview/> (accessed 2016-02-25).
[51] M. Horridge, N. Drummond, J. Goodwin, A. Rector, R. Stevens, H.H. Wang, The
manchester OWL syntax, in: Proceedings of the OWLED⁄06 Workshop on
OWL: Experiences and Directions, Athens, Georgia, USA, 2006.
[52] Apache Jena Fuseki, 2016. <https://jena.apache.org/documentation/fuseki2/>
(accessed 2016-03-29).