Journal of Industrial Information Integration 2 (2016) 1–10

Contents lists available at ScienceDirect

Journal of Industrial Information Integration

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

Ontology-based human–machine integrated design method for

ultra-precision grinding machine spindle
Haibo Hong a, Yuehong Yin b,∗
a
State Key Laboratory of Mechanical System and Vibration, Institute of Robotics, Shanghai Jiao Tong University, Mechanical Building B, Room 1501, 800
Dongchuan Rd., Minhang District, Shanghai 200240, China
b
State Key Laboratory of Mechanical System and Vibration, Institute of Robotics, Shanghai Jiao Tong University, Mechanical Building A, Room 914, 800
Dongchuan Rd., Minhang District, Shanghai 200240, China

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

Article history: This paper proposes an ontology-based design method which integrates human’s knowledge and experi-
Available online 14 April 2016 ence with computer’s inference and computational capabilities for the spindle of ultra-precision grinding
machine. A complete design framework is initiated based on a unify ontology base, which is built to in-
Keywords:
Knowledge integration tegrate human’s experience with computer’s database. The spindle’s bearing and drive type are automat-
Information integration ically selected by defining the experience-based fuzzy inference rules, and applying the similarity-based
Human–machine integrated design instance search method. After the geometric model and finite element model are conducted, the static,
Ontology dynamic and thermodynamic behaviors of the spindle are optimized. Consequently, the design indices of
Spindle the spindle of an ultra-precision grinding machine have been satisfied successfully.
© 2016 Elsevier Inc. All rights reserved.

1. Introduction mechanism, and thermodynamic optimization. Cao and Altintas

Ultra-precision spindle is the core component of large ultra-
precision machine tools, and its performance directly affects the
production efficiency of the machine and the surface quality of the
workpiece. With the increasingly fierce competition in the market,
the precision and efficiency of the machine tools are growing con-
tinually, imposing more and more harsh requirements on the spin-
dle’s performance, especially for the spindles serving optical mirror
manufacturing, where a high speed, a high stiffness, a high accu-
racy and a large required torque need to be guaranteed simultane-
ously. It is quite difficult to accomplish this task purely depending
on either human’s experience or computer’s analyses, due to the
long design lead time, the high cost and the big risk. Therefore,
it has become an urgent problem of how to integrate designers’
knowledge and experience with computer’s powerful computation
and deduction ability, so that the advantages of the both can be
utilized to enhance the efficiency as well as the success rate of de-
signing ultra-precision spindles.
Spindles can be classified into different types according to the
bearing mechanism (e.g., rolling, hydrostatic, aerostatic, etc.) or
transmission mode (gear, belt-pulley, direct-coupling, motorized,
etc.). Up to present, extensive studies have been conducted on
the design and optimization methods of various types of spin-
dles, mainly focusing on the structure design of bearing and drive
∗
Corresponding author.
E-mail addresses: honghb1986@sjtu.edu.cn (H. Hong), yhyin@sjtu.edu.cn (Y. Yin).
[1] simulated the concentric and eccentric forces of a ball bearing
with nonlinear finite element method, and established a dynam-
ical model of the spindle, based on which the natural frequency,
bearing stiffness and deformation of the spindle system were ob-
tained. The design of hydrostatic bearing mainly focuses on journal
bearing [2], thrust bearing [3] and restrictor [4]. Several computer-
aided design tools have also been developed to relieve the design-
ers of the tedious computation burden and to improve design ef-
ficiency. Cavdar [5] designed a computer software system integrat-
ing multi-objective optimization algorithm, which automated the
design process from requirement definition to parameter selection
of the journal hydrostatic bearing. Cheng and Rowe [6] also pre-
sented a design strategy for hydrostatic bearing design to automate
the parameter selection process, including bearing type, configura-
tion, material, etc. Liang et al. [7] proposed an expert system for
hydrostatic bearing design, and divided the total design process
into logical selection and optimization, thus enhancing the design
efficiency. However, a systematic design method for ultra-precision
spindle covering the whole process from customers’ requirements
to structure parameters is far from complete. More importantly,
the existent computer-aided design tools have not effectively intro-
duce designers’ knowledge and experience into the design process,
especially for the most uncertain conceptual design phase. Thus,
there is still a large room for the design efficiency to be improved.
In order to integrate designers’ experience into computer-aided
design, first a unified and unambiguous knowledge representation
and management method needs to be established. Designers’
experience is highly personalized, fuzzy, and hard to be put into

http://dx.doi.org/10.1016/j.jii.2016.04.003
S2452-414X(16)30019-X/© 2016 Elsevier Inc. All rights reserved.
2 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10
Fig. 1. Ontology-based human–machine integrated design framework.
formulas. Moreover, the descriptions of the same knowledge
from various designers may take different forms of language and
vocabulary, resulting in ambiguity and polysemy during knowl-
edge representation. As a representation form with unambiguous
semantics, ontology can eliminate the semantic contradictions
among different knowledge representation forms, thus helping to
build a unified knowledge base [8]. Originating from the philo-
sophical concept of “Ontology”, ontology here is defined as a
specific, formalized and standard description for sharing concep-
tual models. Domain knowledge with specific semantics [9–11] can
be constructed by defining a set of concepts and their semantic
relationships. The key feature of ontology is its reusability and
consistence, based on which the relations among different domain
knowledge can be established via semantic relationships.
The motivation of this paper is to propose an ontology-based
human–machine integrated design method for the spindle of ultra-
precision grinding machines, aiming at fully integrates human’s
knowledge and experience with computer’s powerful inference and
computation capabilities. First, the human–machine integrated de-
sign framework was presented, with a well-structured ontological
knowledge base, which is composed of function, behavior and mor-
phology. Second, in order to realize the automated inference ability
of computers in selecting the bearings and drive types of spindles,
the experience-based fuzzy inference rules and similarity-based in-
stance search method are developed. Third, the static, dynamic and
thermodynamic behaviors of the spindle are optimized after the
geometric model and finite element model are conducted. Finally,
the design indices of the spindle of an ultra-precision grinding ma-
chine have been satisfied successfully.
2. Ontology-based human–machine integrated design
methodology
2.1. Human–machine integrated design framework
The design of ultra-precision spindle is not only a process from
customer requirement analysis to the specification of structural pa-
rameters of all the parts, but also a cognition process of the inter-
action among interdisciplinary knowledge. By combining the ad-
vantages of Pahl and Beitz’s traditional design framework [12] and
Gero’s FBS design prototype [9], a human–machine integrated de-
sign framework is proposed, as shown in Fig. 1.
Originating from requirement (R), four results defined as func-
tion (F), configuration (C), structure (S) and details (D) are gen-
erated by planning and clarification, conceptual design, embodi-
ment design and detailed design respectively. Parameter optimiza-
tion happens at the kinematic (procedures 3, 2 and 3 ), dynamic
(procedures 4, 2 and 4 ) and geometry (procedures 5, 2 and 5 )
levels, when the physical behavior (Bp) is unqualified with the
expected behavior (Be) after comparison (procedure 2). Wherein,
physical behavior is derived from the generated configuration (pro-
cedure 3), structure (procedure 4) and details (procedure 5), mean-
while expected behavior is deducted from the requirement and
function by the computer (procedure 1). Different levels of recon-
figuration happen when local optimization cannot satisfy the de-
sign goal (procedures 6, 6 and 6 ’). Ontological modeling builds
a well-organized architecture of domain knowledge for design, and
provides a normative logical foundation for automatic epistemolog-
ical deduction by computer. Human experience and knowledge are
participated when qualitative information and uncertainties are in-
volved, especially in conceptual design and decision-making during
parameter optimization. Three core stages, including planning and
clarification, conceptual design and parameter optimization, which
determine the effectiveness of the design, are described as follows.
2.1.1. Planning and clarification
This process refers to the conversion from requirements to
functions. The design requirements are usually determined by the
operating conditions defined by the customers. It is essential to set
up a requirement list acting as a guideline during design and a
standard for evaluation. Furthermore, requirement is described as
the spindle’s function and behavior, which act as the direct princi-
ple for the design. The joint discussion of designers and customers
is necessary for the requirement definition process, because clearly
defined requirements would effectively suppress the uncertainties
during subsequent design, thus raising the design efficiency and
success rate. For grinding machine spindle, the operating condi-
tions involved in the design include workpiece material, grinding
feed speed, cutting depth, grinding force, installation form, etc. The
design requirements of the spindle include rotating speed, stiff-
ness, torque, damping, weight, cost, etc. This process will not be
detailed discussed in this paper, as it is relative simple and formal
in spindle design.
 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10
Fig. 2. Ontology base for the design of ultra-precision spindle.
2.1.2. Conceptual design
Conceptual design is a process that determines the configura-
tion according to the required function and behavior. For spin-
dle design, the primary task is to choose a suitable drive form
and bearing type, which largely determines the fundamental per-
formance of the spindle. It is recognized as the most significant
and creativity-requiring phase that has the dominating influence
on the product quality. Conceptual design relies extensively on
human knowledge and experience, especially in creative design
where designer’s intuition plays a central role. Therefore, improv-
ing the design using existent spindles as the reference is extremely
important for enhancing the design efficiency and can avoiding
unnecessary and improper configuration. Finding existing config-
urations satisfying current requirements is particularly important
for novices. Thus, an instance base of spindles and a configuration
search method based on similarity are needed, then, fast searching
and matching of configurations according to the required function
and behavior can be implemented, so that a proper configuration
would be determined.
2.1.3. Parameter optimization
After the configuration of the spindle is determined, structural
parameters of all the parts are optimized to equip the spindle’s
behavior with maximum dynamic stiffness and material remove
rate under the condition of minimum size and energy consump-
tion. The first task in this phase is to build the geometric and finite
element model of the spindle. Then, the static, dynamic and ther-
modynamic analyses according to physical principles can be con-
ducted to optimize the structural parameters.
2.2. Ontological modeling of the knowledge
A unified and unambiguous representation and management of
knowledge are critical to realize the automatic machine reasoning
and interaction among designers and machines during the design.
Ontology base is conducted to reach a consensus on the under-
standing of one specification of ultra-precision spindles. The ontol-
ogy bases contain the knowledge of different structures from var-
ious domains involved in the whole design process. According to
the definition for design ontology proposed by Gruber [8] and Gero
[13], we develop the three types of top level ontologies for ultra-
precision spindle design: function ontology, behavior ontology and
morphology ontology. Each of them can be decomposed into cor-
responding sub-level ontologies, as shown in Fig. 2.
3
2.2.1. Function ontology
Function refers to the expected performance of the product,
and it directly influences design activities, such as problem de-
composition, physical modeling, etc. Establishing the function on-
tology model provides an abstract but straightforward method for
understanding and expressing the spindle’s functions. To ensure
the independency of functions and the generality of understand-
ing, here, we adopt the function basis model with a set of func-
tions (verbs) and flows (nouns) proposed by Hirtz and Stone [14].
A function of a component or assembly can be described us-
ing a verb from the function list, combined with a noun from
the flow list, for example: “glass grinding” can be represented as
“brach::separate::remove::grind” and “material::solid::object::rigid
bogy”. The functions and flows both have three levels: primary,
secondary and tertiary.
2.2.2. Behavior ontology
Behavior refers to the characteristics of the product determined
by the internal factors like structure, material, etc. Each behavior
can be measured by parameter, i.e., specific values with physical
units. Behavior ontology can be categorized into expected behavior
and physical behavior due to the prototype of human–machine in-
tegrated design. The former indicates how the system should per-
form, while the latter characterizes the effective behavior of the
system. Currently, there is no widely applied behavior ontology
due to diverse behaviors of various products. Behavior ontology
can be divided into functional behavior and non-functional behav-
ior. The former describes the embodiment level of some function,
which is usually realized by complex synthesis of multiple com-
ponents, while the latter comes directly form customers’ require-
ments which can be apportioned to the components. For a spin-
dle design, its functional behavior ontologies include load, speed,
stiffness, accuracy, torque, damping and temperature, and the non-
functional behavior ontologies include weight and cost.
2.2.3. Morphology ontology
Morphology is the core part of the spindle’s ontology base, and
it is the knowledge about a product’s assembly form and compo-
nents details. In order to build a unified product knowledge base
in accordance with the design process, morphology ontology can
be divided into three levels: configuration class, structure class and
detail class. Configuration ontology specifically extracts classes en-
capsulating function and their intrinsic attributes. Thus, it is the
representation of all the legal combinations of components as well
as relevant technical and financial constraints. Structure class is the
4 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10
Fig. 3. Ontology base for the design of ultra-precision spindle.
embodying of configuration, which refers to the physical compo-
nents of the spindle.it has a hierarchical structure, and each of such
class can be further decomposed into various sub-classes down to
single parts. Port and energy ontology are defined to represent
the structure ontology, with the semantic relations of “has” and
“use”. Detail class stores the geometric and physical properties of
the component, including shape, topology and topography etc. de-
tail are embodied numerically by variables, including integer type
(number of pockets) and float type (pressure, flow rate, etc.).
Close semantic relationships exist among different classes of
the design ontology. The relation between function and configu-
ration is “synthesis”, i.e., one function can be co-realized by the
synthesis of several configurations, and one configuration can also
realize multiple functions. Physical behavior and morphology has
the relation of “perform”. As the most important variables during
parameter optimization of the spindle, the details of the morphol-
ogy determine the function and behavior that it performs. There
are semantic relations in the internal hierarchical structure of mor-
phology, where structure is the embodiment of configuration, and
then can be detailing into details.
3. Conceptual design of ultra-precision spindles
As is well recognized, conceptual design can be seen as a map-
ping from function and behavior ontologies to configuration on-
tology. There are mainly three kinds of computer-aided conceptual
design methods, i.e., function-based [12], grammar-based [15], and
analogy-based [16] methods. Due to their own merits, these three
methods have been used comprehensively in the conceptual de-
sign of ultra-precision spindles. Similarly, here we divide concep-
tual design process into three phases, i.e., bearing and drive type
selection according to experience-based rules, analogical configura-
tion design based on instance search, and function evaluation and
redesign based on semantic relationships. Thus, a conceptual de-
sign framework with a knowledge searching system is introduced
as shown in Fig. 3. The required input data for the spindle design
requirement, such as the accuracy, stiffness, etc. are entered into
the knowledge searching system, which is established by design
experts and engineers, and is modeled using membership func-
tions. Instance base of spindle is developed according to the mor-
phology ontology. Similarities of every searching keyword are com-
puted with design principles and rules, and then aggregated using
the maximum method. The result of the aggregation is applied to
obtain the most similar instances for reference. Furthermore, func-
tion evaluation is conducted for these instances based on semantic
relations for design modification. Based on this approach, we have
developed a software demo that was applied successfully for the
conceptual design of spindles.
3.1. Selection of bearing and drive type
Proper bearing and drive type should be selected according to
different requirements on the spindle’s behavior. As the key issue
of conceptual design, this process depends on the using experience
of different types of spindles of the domain experts. Due to the
fuzziness of the representation of experience and knowledge, the
specific behavioral requirements need to be fuzzified so as to re-
alize an efficient selection. Besides, the membership relations be-
tween the expected behaviors and the corresponding bearing and
drive behaviors need to be established, and the experience-based
fuzzy inference rules should be made to automate the selection
process of computers.
3.1.1. Fuzzification of behavior ontology
Behavior ontology determines the bearing and drive types. Gen-
erally, various ranges of the behavior are distinguished and de-
scribed empirically, taking the form of fuzzy terminologies such
as “high”, “middle”, and “low”, e.g., hydrostatic bearing has a high
damping and precision, and rolling bearing has a low damping,
high stiffness, etc. As for the spindle, except for damping that can
be defined in the same fuzzy way, other behavior ontologies, such
as precision, stiffness, speed, load, weight, cost, etc., all hold spe-
cific values of requirement, which play an even more important
role during the design.
Different fuzzy terminologies can be distinguished and de-
scribed via defining various ranges of the behavior, i.e., the rela-
tionship between a specific value characterizing a behavior and a
fuzzy terminology can be defined via membership functions (MF).
Taking the speed behavior as an example, for different types of
spindles it is described by the concepts of high speed, middle
speed and low speed. Here, we define the MF of “low speed” as:

1   x≤c
fm (x; c, σ ) = −(x−c )
2 (1)
exp 2σ 2
x>c
where x denotes the required speed, and c is the critical reference
value featuring “low speed” determined by experience. When x is
smaller than c, it falls right into the range of “low speed”, while for
x larger than c we use Gaussian MF to judge how close the cur-
rent speed is to the “low speed”. The detailed shape of the Gaus-
sian MF is determined by σ . If defining the absolutely “low speed”
range as (0, 30 0 0], and the condition when x is greater than 60 0 0
 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10 5

Fig. 4. The definition interface of spindle behavior ontology MFs.

as absolutely not “low speed” (MF value fm < 0.1), then, accord- Table 1
Inference rules for the bearing types of ultra-precision spindles.
ing to the requirements, corresponding parameters can be chosen
as: c = 30 0 0, σ = 1398. Similarly, the MFs for “middle speed” and Input Output
“high speed” can be defined. It should be noted that the selec- Speed Stiffness Accuracy Damping Life Temperature Cost Bearings type
tion of parameters c and σ relies on the experience of the ex-
perts and designers of this domain, therefore, different behavior H/M/L∗ H/M H/M L L M L/M Rolling
L H/M L H H H M/H Hydrostatic
ontologies may also have MFs of various types, e.g., Bell-shaped
L/M M H/M H M H M Hydrodynamic
MF, or sigmoidal-shaped MF. Fig. 4 shows the MF defining inter- H L L L H L M Aerostatic
face for the behavior ontology in the conceptual design software, H M L M H L H Magnetic
including the definitions of the MFs types and the corresponding ∗
H: High, M: Middle, L: Low.
curve shapes. The parameters of the MFs are set by referring to
previous researches [17] and the author’s rich experience in ultra-
precision machine tool design. The major behavior ontologies of
ultra-precision spindle are described by Gaussian MFs, so there are effectiveness ratio. By comparison with rolling bearings, hydro-
totally eight parameters to define. static and hydrodynamic bearings have the advantage of achieving
As for the behavior ontology “damping” of the spindle, it can- minimum radial and axial runout, and can achieve good damping
not be defined directly with specific values in the behavior re- and high stiffness. However, there is a limit on the speeds, due to
quirements, thus it is often described by the fuzzy terminologies the increased heating of the fluid with internal shearing effects.
of “low”, “middle” and “high”. To ensure the consistence of the rep- Besides, hydrodynamic bearings are used only where operation
resentation of behavior ontology, we define these three terminolo- conditions are in the purely fluid friction range without speed vari-
gies in a discrete manner as “0”, “0.5” and “1”, respectively, and ations. Working on the same principle as liquid-lubricated bear-
the distance between adjacent levels is 0.5, with the correspond- ings, aerostatic bearings are employed when much higher rota-
ing membership grade of 0.5. For example, if the requirement of tional speeds need to be achieved. The active medium is gaseous
spindle damping is “high”, while that of bearing damping is “low”, air, with a lower viscosity than that of the liquids by two to three
then the distance between them is 1, and the membership grade orders of magnitude. For magnetic bearings, the friction can be ne-
is 0. glected because of the relatively large air gaps between the rotor
and stator. They have the maximum possible static stiffness but
3.1.2. Experience-based inference rules the load rating is significantly low than the other bearings. Due to
The inference rules concerning the selection of spindle bearing the high costs of a complex control system and peripherals these
and drive type are made according to the experience of domain spindles are currently used only in special applications. Therefore,
experts. If taking the behaviors derived from the requirements as the experience based inference rules are proposed as shown in
the input, and the types of bearing and drive as the output, then Table 1. For each possible output of bearings type, required in-
the relationship between input and output can be determined, and put conditions of the corresponding fuzzy terminologies of behav-
the inference rule corresponding to each output can be set. ior ontologies are determined.
Bearing types of the spindle include rolling, hydrostatic, hy- It can be inferred from the MFs shown in Fig. 4, the specific
drodynamic, aerostatic, magnetic, etc., which are chosen according behavior requirement could fall in a “transitional” interval, i.e., it
to the joint factors including speed, stiffness, precision, damping, might be in two definition ranges (fm > 0.1) simultaneously, lead-
temperature and cost. According to experts’ experience and previ- ing to multiple inference results under the same input conditions.
ous researches [18], spindles with high axial and radial stiffness Consequently, to attain a more appropriate result by a more spe-
are generally mounted on rolling bearings. High-precision spin- cific inference process, we evaluate the similarity between the be-
dle bearings combine good radial run-out and stiffness proper- havior requirement and the experience-based reference conditions
ties with low assembly and maintenance effort, and a good cost- via calculating the MFs of each behavior ontology:
6 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10

Table 2
Type selection results of the ultra-precision spindle.

Output Input

Speed Stiffness Accuracy Damping Life (w5 = 0.05) Temperature Cost (w7 = 0.1) Total similarity
(w1 = 0.2) (w2 = 0.3) (w3 = 0.2) (w4 = 0.1) (w6 = 0.05)

Hydrostatic 0.7776 1 0.9098 1 0.0 0 01 0.0 0 08 1 0.8375

Rolling 1 1 0.2204 0 0.2262 0.2739 1 0.6691
Hydrodynamic 0.7776 0.5610 0.2204 1 0.9113 0.0 0 08 1 0.6135
Magnetic 0.0 0 01 0.5610 0.9098 0.5 0.0 0 01 0.8660 0.0 0 01 0.4436
Aerostatic 0.0 0 01 0.0 0 01 0.9098 0 0.0 0 01 0.8660 1 0.3253


n are denoted as fm (xi ):
Sb = ωi f m ( xi )
i=1
where Sb ∈ (0.1] denotes the total similarity of the behavior, and
ωi ∈ [0.1] are the weight factors describing the significance levels
of the behaviors specified by the designer, with the summation of
the weights being unity. A larger Sb indicates a larger probability
that the output is consistent with the design requirements. It is
worth noting that when the search conditions include any default
value (i.e., no requirement is specified for the corresponding be-
havior ontology), we consider that the MFs between the target be-
havior and the concepts of this behavior ontology are all “1”, but
their weights are all “0”.
For the bearings type of the ultra-precision spindle
in this paper, the input conditions are: speed = 40 0 0 rpm
(w1 = 0.2), stiffness = 400 N/μm (w2 = 0.3), accuracy = 0.6 μm
(w3 = 0.2), damping = high (w4 = 0.1), life = 90 0 0 h (w5 = 0.05),
temperature = 25 °C (w6 = 0.05), cost = 20,0 0 0 dollars (w7 = 0.1).
The similarity results are computed via the proposed inference
algorithm, as shown in Table 2:
Table 2 indicates that the choices according to the similarity
in a descending order are hydrostatic spindle, roller spindle, hy-
drodynamic spindle, magnetic spindle and aerostatic spindle, re-
spectively. We set that the bearing type fails to match the current
behavior requirement when the total similarity is less than 0.5.
Thus, the candidate types would be hydrostatic, roller and hydro-
dynamic bearings, wherein hydrostatic bearing is the best choice
due to its much larger similarity. As is determined by speed and
torque, the driving type can also be selected by setting correspond-
ing experience-based inference rules for every output.
3.2. Analogical configuration design based on instance searching
The task after the selection of bearing and drive types is to de-
termine the fundamental configuration of the spindle, which is im-
plemented by the analogical method of searching existent spindle
instances. These instances are derived from the internet search and
provided by the manufacturer. According to the definition of ontol-
ogy base, each instance is modeled according to the correspond-
ing features of function, behavior and structure. Thus, by compar-
ing the similarity grade between current requirements on behavior
and structure and the behavior and structure ontologies of existent
instances, the basic configuration can be determined in an easier
way. The similarity grade can be expressed by:

n1 
n2
Si = Sb · Sd · ωi f m ( xi ) · gs ( b j )
i=1 j=1
where the total similarity is denoted by Si ∈ [0.1], Sb and Sd are the
MFs of the inference results of bearings and drive types, n1 and n2
are the input numbers of numerical (behavior ontology) and string
(structure ontology) type requirements, respectively. The Gaussian
MFs for computing the similarity of numerical type requirements
(2)  
− (x − c )
2
fm (x; c, σ ) = exp (4)
2σ 2
where x is the required target value of the behavior ontology, c is
the parameter value of the same kind of behavior ontology of the
analogical instance, and σ determines the MF shape.
In Eq. (3), gs (bj ) is the Boolean function specifying the similarity
of string type requirements, and it is defined as:
1, if b j is not specified or equal
gs ( b j ) = (5)
0, if b j is specified but not equal
where bj denotes the string type requirement, which usually can-
not be expressed by specific values, and it presents the features of
“yes” and “no” or the matching of specific terms. When comput-
ing the string type similarities, instances with inconsistent require-
ments would be defined directly as not similar (gs (bj ) = 0). Refer-
ring to Eq. (3), the final similarity of the instance is zero only if
any one of the string type requirements is not consistent. There-
fore, string type similarity is determinant for the overall similarity,
and it is used only when particular requirements are imposed on
the spindle’s structure, otherwise many potential instances may be
rejected. During the instance search process, we only specify the
Boolean functions for the bearings type and drive type, i.e., we take
the type set whose similarity is greater than 0.5 computed in Sec-
tion III as the structure requirement, indicating that the instances
out of this set would not be taken into account.
For the spindle in this paper, the requirements of behavior on-
tology and structure ontology determined by operation conditions
are input into the instance search interface of the conceptual de-
sign system, and the spindle instances (only hydrostatic, hydrody-
namic and rolling gs (bj ) = 1) in the ontology base are compared
and searched, as shown in Fig. 5. A larger similarity indicates a
higher reference value of the instance for configuration design.
When Si < 0.5, the instance would be rejected due to its remark-
able differences.
3.3. Function evaluation of the spindle
The structure of the product must satisfy all the function re-
quirements according to the basic principle of product design. The
above search process for the basic configuration takes the behavior
ontology of the spindle as the guiding rule. Thus, we need to make
sure whether the function requirements are fully satisfied before
parameter optimization. It can be noted from Fig. 2 that close se-
mantic relations exist between the structure ontology and function
(3)
ontology, then, the functional verification of the instance can be
implemented via the matching between morphology and function
requirements based on semantic search, in order to judge whether
any function has not been realized. If there does exist any unreal-
ized function, a redesign is needed for that function. Fig. 6 shows
the morphology inquiry and functional verification interface of the
conceptual design system for the instance, and the picture of the
 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10 7

Fig. 5. Instance search results of ultra-precision spindle.

Fig. 6. Structure ontology and functional realization of the spindle instance.

spindle as well as the resource linking provide the designers with
more detailed information.
4. Parameter optimization of the ultra-precision spindle
Following the determination of the configuration, the core task
of parameter optimization is to establish the relationship between
feature ontology of the structure and behavior ontology, where the
feature ontology includes the dimensions of spindle axis, the po-
sition of bearing and its structures, etc. There are three types of
behavior ontology: static, dynamic and thermodynamic, mainly in-
cluding deformation, stiffness, mode, thermal error, etc. On the
whole, the parameter optimization of the spindle consists of two
parts: dynamic optimization and thermodynamic optimization.
4.1. Dynamic optimization
Fig. 7 shows the directly driven hydrostatic spindle model gen-
erated by the conceptual design (Section 3). The spindle is directly
driven by the motor, and the grinding wheel is arranged at the
other end of the spindle by a conical shaped surface. The spindle is
supported by two radial hydrostatic bearings and one axial thrust
bearing. Here, constant pressure oil supply is applied, and the oil
fills the pocket by going through the orifice restrictor from the en-
try, then goes back to the oil tank through the return hole. An oil
film is formed between the bearings and the spindle and props up
the spindle separating the bearing sleeves.
A finite element model is a prerequisite for dynamic optimiza-
tion, based on which the dynamic model of the spindle can be
established with Euler beam theory [19]. As shown in Fig. 8, the
spindle is divided into several beam units according to cross sec-
tion shape and length, and the junction of two connected beam
elements forms a node. Assuming the spindle and the grinding
wheel are rigidly connected, the spindle is divided into 12 beam
elements and the finite element model has 13 nodes in total. Hy-
drostatic bearings are simplified as compressive springs and tor-
sional springs, whose stiffness is computed via the stiffness model
of hydrostatic bearing. According to the design principles of hydro-
static bearings, the stiffness equations for radial and thrust bear-
ings are:
6Ae Ps (β − 1 )cosθ 6Aez Ps (β − 1 )
kr = , ka = (6)
h0 β 2 hz0 β 2
8 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10

Fig. 7. Schematic of the spindle’s assembly. 1: Rotor; 2: stator; 3: the spindle box; 4: radial hydrostatic bearing; 5: spindle; 6: axial hydrostatic bearing; 7: wheel.

Fig. 8. The finite element model of the spindle.

⎡ ⎤
where Ae and Aez are the effective support areas of the radial and ω12 0 ··· 0
thrust bearings, and hz and hz 0 are the clearances of these two ⎢0 ω22 ··· 0 ⎥
kinds of bearings, respectively. Ps is the oil supply pressure, β is [] = ⎢ . .. ⎥ (11)
⎣.. ..
. ⎦
the restrictor ratio, and θ is the half of the circumferential angle . 0
of the radial bearing’s pocket. 0 ··· 0 ω5k
2

The deformation of the operating spindle caused by external and ωj (j = 1, …, 5k) is the natural frequency of the spindle assem-
forces is comprised of the displacements from five directions, i.e., bly. The damping matrix is
the axial displacements (u, v, w) along x, y, and z direction and the ⎡ ⎤
rotary displacements (θ x , θ y ) about x and y axes. According to the 2ς1 ω1 0 ··· 0
Lagrange equation ⎢0 2ς2 ω2 ··· 0 ⎥
[B] = ⎢. .. ⎥ (12)
  ⎣.. .
..
. 0
⎦
d ∂ Ek ∂ Ep 0 ··· 0 2 ς5k ω5k
+ = f (7)
dt ∂ q˙ ∂q
and ζ i is the damping coefficient of the spindle. The variable
η can be computed by solving the 2nd order differential equa-
and combine the spindle damping, the finite element model of the
tion, the generalized variable {Q} can be obtained by solving
spindle can be expressed as:
{Q} = []{η}([] is the regular transform matrix.
The goal of the dynamic optimization is to attain a higher low-
[M]{Q̈ } + [C]{Q˙ } + [K]{Q } = {F } (8) order natural frequency and a smaller deformation via tuning the
parameters of feature ontology. This is a typical multi-objective op-
wherein the generalized variable vector is {Q} = [u1 v1 θ x 1 θ y 1 w1 timization problem, whose variables are the bearing length L1 and
… uk vk xk yk wk ]T, [M] and [K] are the mass matrix and stiffness the distance L2 between the two bearings. The objective function
matrix, respectively, and is:
f (L1 , L2 ) = fz (L1 , L2 )/ fz0 + δ (L1 , L2 )/δ 0 (13)

k 
k 
l
[M] = [m]i , [K] = [K]i + [K]ci · [m]i (9) where fz is the 1st order natural frequency of the spindle assem-
i=1 i=1 i=1 bly, δ is the maximum deformation under the external force at the
spindle’s extreme, and fz 0 and δ 0 are the values of these two terms
[m]i is the mass matrix for beam element i, [K]i is the corre- at the initial condition.
sponding stiffness matrix, while [K]ci (k = 13, l = 5) is the stiffness During the optimization iteration, constrained random direction
matrix of the spring between the nodes. {F} is the force matrix in- search method is applied to obtain fz 0 and δ 0 . Then, it is iterated
cluding the node forces and the resultant force on the spindle. to solve the minimum value of the objective function. The natu-
By decoupling the finite element model, the natural frequency ral frequencies of first five orders before and after optimization
and the deformation of the spindle can be obtained. The dynamical are 218.1(358.8), 218.1(358.8), 514(514), 763.9 (1389.2) and 763.9
model after decoupling is (1389.2). Fig. 9 shows the deformations at node 12 before and af-
ter optimization. It can be seen that the natural frequencies of first
{η̈} + [B]{η}
˙ + []{η} = {N} (10) two orders are raised by 64.5%, and that of fourth and fifth orders
are raised by 81.8%, meanwhile the deformation is decreased by
where the stiffness matrix is 47.5%. After optimization, the stiffness of the hydrostatic bearings
 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10
Fig. 9. The deformations of node 12 before and after optimization.
is enhanced by 3 times, and the natural frequencies of the spindle
assembly have been considerably improved.
4.2. Thermodynamic optimization
The thermal deformation error is a key factor affecting the spin-
dle’s precision, stiffness and life expectation, so the thermal analy-
ses and optimization of the spindle play a significant role during
the design process. First of all, the heat source and heat trans-
fer conditions of the spindle should be calculated, and the tem-
perature field distribution and the deformation of the spindle can
be analyzed so as to propose a strategy that reduces the thermal
deformation.
4.2.1. Heat source calculation
The heat sources of the spindle mainly include the direct-driven
high speed asynchronous motor and the friction between bearings
and oil. Due to the face that the friction between bearings and oil
is very small, and the temperature of the oil is maintained constant
through an oil cooling system removing the heat produced by fric-
tion. Thus, during practical analysis, we only consider the motor’s
dissipation heat. The dissipation can be classified into mechanical
dissipation, electrical dissipation, magnetic dissipation, additional
dissipation, etc., where the former three are the main types, while
the additional dissipation only accounts for 1–5% of the total rated
power [20]. The heat Qf of the spindle motor can be expressed by
the following empirical equation
Mm n 1 − η
Q f = Nm (1 − η ) = · (14)
9550 η the design of a hydrostatic spindle on an ultra-precision grinding
where η is the efficiency, Nm is the input power, Mm is the output
torque, and n is the motor speed.
4.2.2. Calculation of the heat transfer condition
In general, heat transfer proceeds in three ways, conduction,
convection and radiation, and it is a complex process in a machine
tool. For a hydrostatic spindle system, convection and conduction
are the primary means, including the heat conduction of between
the metallic bearing parts, and the heat convection between the
bearing, air and oil.
The body material of spindle is alloy steel (type Cr40), and the
bearings are made of tin bronze. The heat conducting carriers of
them are both free electrons. According to the free electron theory,
the electrons in metal can be regarded as free electron gas. There-
fore, the heat conduction rate of metal can be computed with the
heat conduction rate formula of ideal gas
 
1 π2 kT π 2 ne k2 T
λe = k · ne ve le = τe
3 2 Ee0 3m
9
where k is the state density of a unit volume, ve is the speed of
electrons, τ e is the relaxation time of electrons, ne is the number
of electrons in a unit volume, m is the mass of electron, T is the
temperature, Ee0 is the free electron energy of a unit volume, and
le is the mean free path of electron.
The convection heat transfer coefficient between bearing and
air is h = Nuλfluid /d, wherein Nu is the Nusselt number, and λfluid
is the heat conduction rate of air. The characteristic dimension of
the contact area is denoted as d, and for a cylinder it is the di-
ameter, while for convection between two planes it is the dis-
tance between them. The convection heat transfer coefficient be-
tween the bearing and oil is computed with the empirical formula
h0 = λoil Pru/(3vxl ), where λoil is the heat conduction rate of the oil,
u is the convection speed, v is the kinematic viscosity of the oil
under operating temperature, and xl is the characteristic length of
bearing.
4.2.3. The temperature field distribution of the spindle
After simplifying the geometric model of the spindle, it is im-
ported into the ANSYS software to implement the configuration of
material property, meshing, the configuration of boundary condi-
tions and thermal load, in order to calculate the final temperature
distribution of the spindle. Due to the symmetric feature, we can
simplify the calculation by taking a quarter of the whole spindle.
The results are shown in Fig. 10.
4.2.4. Thermodynamic optimization
The thermodynamic optimization of the spindle can be imple-
mented from three aspects: increasing the distance between heat
source and the output end of the spindle, reduce the size of con-
tact area, and adding a cooling system. First, for the spindle struc-
ture, a postposition drive motor is adopted to make the motor heat
source far from the spindle’s front end to weaken its influence on
the spindle and facilitate cooling. Second, the temperature rise of
the spindle’s shell is mainly due to the heat conduction between
the shell and the motor’s stator, so it is necessary to reduce the
contact area between them. Fig. 10(a) and (b) shows the temper-
ature fields of the spindle under different contact areas. Third, as
the major heat source of the spindle, the motor produces a lot of
heat during its operation. Thus, a circulating water cooling system
for the motor can be added, so that the cooling water would ab-
sorb most of the heat to lower the temperature of the motor, as
shown in Fig. 10(c).
5. Implementation
The parameter optimization has been applied successfully to
machine for large optical mirrors. The machine and spindle must
hold ultra-high stiffness and precision to realize a surface rough-
ness of 3–8 μm, and a subsurface damage layer with thickness
below 10 μm. The appearance and assembly form of the spindle
is shown in Fig. 11(a), with a speed of 40 0 0 rpm, a precision of
0.5 μm, and a torque of 45 N m. Fig. 11(b) shows the silicon carbide
(SiC) workpiece for the grinder, with a diameter of 250 mm. The
grinding results show that the performance of the spindle is satis-
factory, validating the effectiveness of the proposed design method.
6. Conclusion
This paper proposed an ontology-based design method for spin-
dles, which has been applied successfully to the design of the hy-
drostatic spindle on an ultra-precision grinding machine. This ap-
proach integrates effectively human’s knowledge and experience
with computer’s inference and computational ability, thus enhanc-
ing the design efficiency and rationality of spindle design. The con-
(15)
clusions are made as follows:
10 H. Hong, Y. Yin / Journal of Industrial Information Integration 2 (2016) 1–10
Fig. 10. Hydrostatic spindle temperature distribution in different conditions. (a) Large contact area; (b) small contact area; (c) with the effect of a water cooling system.
Fig. 11. The deformations of node 12 before and after optimization. (a) Ultra-precision grinding machine, (b) hydrostatic spindle, (c) SiC workpiece.
(1) This approach proposed a human–machine integrated design
framework, which takes the advantages of both human’s ex-
perience and knowledge with computer’s computation and
logical deduction capabilities. Furthermore, a design ontol-
ogy base composed of function, behavior and morphology
ontologies is established, which unifies the knowledge rep-
resentation forms of human and computer, so that an un-
ambiguous knowledge management method is provided for
integrated design.
(2) This approach built fuzzy reference rules based on experts’
experience, thus realizing the automated selection of bear-
ings and drive types. It provides an instance search method
based on similarity comparison, which helps to determine
the basic configuration of the spindle. In addition, a function
verification method based on semantic search is proposed,
guaranteeing the effectiveness of conceptual design.
(3) The parameter optimization of the behavior ontologies of
the hydrostatic spindle is realized via dynamic optimization
and thermodynamic optimization, which greatly improves
the performance of the spindle.
Acknowledgment
This work is supported by the National Basic Research Program
of China (973 Program) (No. 2011CB013203).
References
[1] Y. Cao, Y. Altintas, A general method for the modeling of spindle-bearing sys-
tems, ASME J. Mech. Des. 126 (6) (2004) 1089–1104.
[2] A. Bouzidane, M. Thomas, Equivalent stiffness and damping investigation of a
hydrostatic journal bearing, Tribol. Trans. 50 (2) (2007) 257–267.
[3] O.-J. Bakker, R.-A. Van Ostayen, Recess depth optimization for rotating, annular,
and circular recess hydrostatic thrust bearings, ASME J. Tribol. 132 (1) (2010)
011103–011109.
[4] X. Zuo, J. Wang, Z. Yin, S. Li, Comparative performance analysis of conical hy-
drostatic bearings compensated by variable slot and fixed slot, Tribol. Int. 66
(2013) 83–92.
[5] K. Cavdar, Computer aided optimization of the hydrostatic radial journal bear-
ings, Ind. Lubr. Tribol. 58 (3) (2006) 129–134.
[6] K. Cheng, W.-B. Rowe, A selection strategy for the design of externally pressur-
ized journal bearings, Tribol. Int. 28 (7) (1995) 465–474.
[7] Y. Liang, W. Chen, Y. Sun, N. Yu, P. Zhang, H. Liu, An expert system for hy-
dro/aero-static spindle design used in ultra precision machine tool, Robot.
Computer-Integr. Manuf. 30 (2) (2014) 107–113.
[8] T.-R. Gruber, Toward principles for the design of ontologies used for knowledge
sharing? Int. J. Human-Comput. Stud. 43 (5) (1995) 907–928.
[9] J.-S. Gero, Design prototypes: a knowledge representation schema for design,
AI Mag. 11 (4) (1990) 26–36.
[10] R. Roy, C.-I.-V. Kerr, P.-J. Sackett, J. Corbett, Design requirements management
using an ontological framework, CIRP Annals-Manuf. Technol. 54 (1) (2005)
109–112.
[11] K.-Y. Kim, D.-G. Manley, H. Yang, Ontology-based assembly design and infor-
mation sharing for collaborative product development, Computer-Aided Des.
38 (12) (2006) 1233–1250.
[12] G. Pahl, W. Beitz, Engineering Design: A Systematic Approach, Springer Science
& Business Media, UK, 2013 (Chapter 6).
[13] J.-S. Gero, U. Kannengiesser, A function–behavior–structure ontology of
processes, Artif. Intell. Eng. Des. Anal. Manuf. 21 (04) (2007) 379–
391.
[14] J. Hirtz, R.-B. Stone, D.-A. McAdams, S. Szykman, K.-L. Wood, A functional basis
for engineering design: reconciling and evolving previous efforts, Res. Eng. Des.
13 (2) (2002) 65–82.
[15] J. Patel, M.-I. Campbell, An approach to automate and optimize concept gener-
ation of sheet metal parts by topological and parametric decoupling, ASME J.
Mech. Des. 132 (5) (2010) 051001–051011.
[16] B.-T. Christensen, C.-D. Schunn, The relationship of analogical distance to ana-
logical function and preinventive structure: the case of engineering design,
Mem. Cogn. 35 (1) (2007) 29–38.
[17] O. Maeda, Y. Cao, Y. Altintas, Expert spindle design system, Int. J. Mach. Tools
Manuf. 45 (4) (2005) 537–548.
[18] E. Abele, Y. Altintas, C. Brecher, Machine tool spindle units, CIRP Annals-Manuf.
Technol. 59 (2) (2010) 781–802.
[19] R. Katz, C.-W. Lee, A.-G. Ulsoy, R.-A. Scott, The dynamic response of a ro-
tating shaft subject to a moving load, J. Sound Vib. 122 (1) (1988) 131–
148.
[20] Y.-C. Shin, Bearing nonlinearity and stability analysis in high speed machining,
ASME J. Eng. Ind. 114 (1) (1992) 23–30.