applied

sciences
Article
Reliability-Based Multi-Objective Optimization Design of a
Compliant Feed Drive Mechanism for Micromachining
Van-Khien Nguyen 1,2 , Huy-Tuan Pham 1, * , Huy-Hoang Pham 3 , Quang-Khoa Dang 1 and Pham Son Minh 1

1 Faculty of Mechanical Engineering, Ho Chi Minh City University of Technology and Education,
Ho Chi Minh City 790000, Vietnam
2 Faculty of Mechanical Engineering, Nam Sai Gon Polytechnic College, Ho Chi Minh City 790000, Vietnam
3 Faculty of Mechanical Engineering, Ho Chi Minh City University of Food Industry,
Ho Chi Minh City 790000, Vietnam
* Correspondence: phtuan@[Link]

Featured Application: micromachining, feed drive mechanisms.

Abstract: In precision engineering, the use of compliant mechanisms (CMs) in positioning devices
has recently bloomed. However, during the course of their development, beginning from conceptual
design through to the finished instrument based on a regular optimization process, many obstacles
still need to be overcome, since the optimal solutions often lie on constrained boundaries or at
the margin of safe/unsafe domains. Accordingly, if uncertainty occurs during the fabrication or
operation of the mechanism, it might lose its functions, rendering the design infeasible. This paper
proposes a universal design process for positioning CMs, consisting of two steps: optimal design of
the pseudo-rigid-body model, and reliability-integrated multi-objective optimization design using
NSGA-II algorithms. This optimization algorithm is applied in the design of a feed drive mechanism
for micromachining. The optimal design is also fabricated and tested. The results calculated for
the displacement amplification ratio, natural frequency, and input/output stiffness using different
approaches, including analytical methods, simulations, and experiments, were compared to evaluate
Citation: Nguyen, V.-K.; Pham, H.-T.;
the efficiency of the proposed synthesis method, and show discrepancies of less than 5%. Thus, the
Pham, H.-H.; Dang, Q.-K.; Minh, P.S. results convincingly support the applicability of the proposed optimization algorithm for the design
Reliability-Based Multi-Objective of other precision-positioning CMs prone to failure in vulnerable conditions.
Optimization Design of a Compliant
Feed Drive Mechanism for Keywords: compliant mechanism; feed drive mechanism; NSGA-II algorithm; reliability-based
Micromachining. Appl. Sci. 2023, 13, optimization; FEM; FORM
4968. [Link]
app13084968

Academic Editors: Shahrokh Hatefi

and Khaled Abou-El-Hossein 1. Introduction
Fully compliant mechanisms (CMs) are monolithic structures that can transmit motion,
Received: 27 February 2023
forces, and torques through the elastic deformation of flexible joints. Unlike traditional rigid
Revised: 7 April 2023
mechanisms, they have no dynamic joints, no clearance between joints, and, therefore, no
Accepted: 13 April 2023
Published: 15 April 2023
friction. Their monolithic nature gives them the spectacular characteristics of high precision
and simple design. Consequently, machining costs and assembly effort are reduced [1].
In recent decades, CMs have gained considerable attention for use in a broad range of
applications in precision engineering, such as in the positioning mechanisms in machining,
Copyright: © 2023 by the authors. high-precision feed drive mechanisms [2,3], fast tool servos [4], nanopositioners for micro-
Licensee MDPI, Basel, Switzerland. scopes or other applications [5], grippers for micro-object manipulation [6,7], constant-force
This article is an open access article mechanisms for force regulation [8], and high-precision hydraulic valve control systems [9].
distributed under the terms and Typically, the approaches commonly used to cope with the diversity of CM configura-
conditions of the Creative Commons tions in design and synthesis can be grouped into three categories, namely, static/kinetic
Attribution (CC BY) license (https:// mathematical model-based approaches [1,10,11], building block approaches [12–14], and
[Link]/licenses/by/ topology, size, and shape optimization-based approaches [8,15–17].
4.0/).

Appl. Sci. 2023, 13, 4968. [Link] [Link]

 Appl. Sci. 2023, 13, 4968 2 of 29

Appl. Sci. 2023, 13, 4968 2 of 27

block approaches [12–14], and topology, size, and shape optimization-based approaches
[8,15–17].
Amongthe
Among thevarious
various applications
applications of CMs,
of CMs, theirtheir use
use as as high-precision
high-precision positioning
positioning mech-
mechanisms has received the most interest [18]. Researchers in
anisms has received the most interest [18]. Researchers in this field usually use bridge-this field usually use
bridge-type [19], lever-type, L-shape levers, and half-bridge [18] or
type [19], lever-type, L-shape levers, and half-bridge [18] or their combination [20] to their combination [20]
to design
design the motion
the motion transmission
transmission platforms.
platforms. Figure Figure 1 illustrates
1 illustrates two examples
two examples of posi-
of positioning
tioning
CMs CMs
using using multistage
multistage displacementdisplacement
amplifiers,amplifiers,
wherein rigidwherein
links rigid links and
and flexure flexure
hinges are
hinges arein
integrated integrated in their configurations.
their configurations. Shape and size Shape and size optimization
optimization approaches approaches
are often usedare
often used as part of the design methodology for this type of structure.
as part of the design methodology for this type of structure. Previous studies have usually Previous studies
have usually
focused on the focused on the
dimension ofdimension
the hingesoftothe hinges tothe
maximize maximize the amplification
amplification ratio.
ratio. Although
Although
the the shape
shape and size ofandthesize of the
elastic elastic
joints arejoints are essential
essential factors infactors
assuringin assuring the defor-
the deformation
mation capacity,
capacity, the length theoflength of the
the rigid rigid
links links determines
determines the displacement
the displacement range, parasitic
range, parasitic motion,
motion,
and and amplification
amplification magnitude. magnitude.
Selecting Selecting
adequate adequate design parameters
design parameters for specificfor specific
purposes
ispurposes is a persistently
a persistently challenging challenging
task. task.

[Link]-rigid-body
Figure Pseudo-rigid-body of
of (a)
(a) aacompliant
compliant 2-DOF
2-DOF stage
stage for
forlocating
locating biomaterial
biomaterial samples
samples [21],
[21],
and (b) a large-stroke nanopositioning stage [22].
and (b) a large-stroke nanopositioning stage [22].

Due
Duetotothe
themultitude
multitudeof ofparameters,
parameters,in inearlier
earlierstudies
studiesthetheselection
selectionofofdesign
designvariables
variables
(DVs)
(DVs) to formulate the problem has often been dependent on the experience of
to formulate the problem has often been dependent on the experience of the
the partic-
partic-
ular
ular researchers
researchersinvolved.
involved. InIn the
the optimization
optimization process,
process, itit is
is also
also often
often assumed
assumed that
that the
the
dimensions
dimensionsofofthethedesign
designparameters,
parameters, as as
well as the
well operating
as the operating conditions of the
conditions of mechanism
the mecha-
after
nismfabrication, are ideal.
after fabrication, areFor applications
ideal. where high
For applications accuracy
where is required,
high accuracy any reckless
is required, any
variation in the fabricated structure, compared with the theoretical design, may result in an
unignorable impact on the performance of the instrument. Practices in precision engineer-
Appl. Sci. 2023, 13, 4968 3 of 27

ing also show that in many realistic input parameters, there is random fluctuation according
to a specific probability distribution function. Examples of the problematic uncertainty
in parameters include fluctuations in cutting forces, inhomogeneous material properties,
tolerance of fabricated parts, assembly accuracy, as well as the imperfect constraints of
the mechanism. This consequently leads to inconsistency in the output performance of
the designed instrument. In some unexpected scenarios, this may exceed the predeter-
mined allowable limit. The likelihood for this jeopardy to occur is called the unsafety or
failure probability of the structure. Determining the failure probability of a system, due to
uncertain external factors, is referred to as the reliability design problem [23].
For reliability-based design optimization (RBDO) problem solving, two calculation
cycles are required: the optimization cycle and the reliability evaluation cycle. Therefore,
the computational cost is one of the major concerns. The most commonly used methods
to solve reliability analysis problems are the first-order reliability method, second-order
reliability method [24], Monte Carlo simulation method [25], double loop method, decouple
double loop method, and single loop deterministic method [26]. While reliability analysis
optimization for CMs has also recently received interest, there is still little research on
this topic [27–29]. In these studies, problems are usually treated via explicit mathematical
functions [30]. The Hasofer–Lind approximation method is appropriate for linear functions,
but not suitable for high-order functions. The Monte Carlo simulation (MCS) method [31],
which evaluates the limit state function of the data samples, is one of the most popular
reliability analysis methods. It is simple to use, and the results are easy to analyze. However,
it often requires many samples to ensure accuracy, leading to high computational costs.
The first-order reliability method (FORM) is based on the Taylor series expansion of the
limit state function, whose approximation may not be accurate for complex nonlinear limit
state functions. Its applicability may also not be appropriate for discontinuous or multiple
failure modes limit state functions. However, it can be integrated into a FEM tool to quickly
evaluate any high-order nonlinear limit state functions, regardless of the need for an explicit
mathematical model. Therefore, the FORM can overcome the drawbacks of the previous
methods, and still effectively solve any complicated reliability-based problems. RBDO of
implicit CM design problems that are difficult to mathematically formulate has not yet
been extensively investigated. Thus, this paper aims to fill this gap.
Regarding the optimal design algorithm used for synthesizing positioning CMs, in
previous research, there has often been a distinction between the study of rigid links and
the design of the elastic components. This paper proposes a reliability-based optimization
method for their design that first optimizes a pseudo-rigid-body model (PRBM) via an
analytical approach. The 2D model is then built in the Ansys ADPL language for the
analysis of its behaviors by the finite element method. Subsequently, a multi-objective
design optimization problem is solved using the nondominated sorting genetic algorithm
II (NSGA-II). This research also integrates a reliability analysis task using the FORM into
the design process to assess the impact of uncertainty in the practical input parameters, and
implements this as an RBDO problem. In this stage, the constraints of the multi-objective
design optimization problem are treated as limit state functions to evaluate the reliability
of the obtained optimal solutions. The above combined design method is applied to design
a compliant micropositioning feed drive mechanism (CFDM). In this paper, its natural
frequency and displacement functions are also analyzed using the PRBM. Eventually,
the optimal design is machined by wire EDM, with experiments conducted to evaluate
the fabricated prototype. The output displacement is tested in a LabVIEW 2014 closed-
loop control system, whose PID parameters have previously been determined using a
single-objective genetic algorithm (GA) in MATLAB 2014a.

2. Design Methodology
2.1. Design Concept
Piezoelectric (PZT) actuators are essential for achieving precise and accurate motion
control in various applications. However, the maximum displacement of these actuators
Appl. Sci. 2023, 13, 4968 4 of 29

2. Design Methodology
Appl. Sci. 2023, 13, 4968 2.1. Design Concept 4 of 27
Piezoelectric (PZT) actuators are essential for achieving precise and accurate motion
control in various applications. However, the maximum displacement of these actuators
is typically
is typically limited
limited to to the
the micrometer
micrometer range. range. ByBy using
using CMsCMs to to transmit
transmit andand amplify
amplify their
their
driving motion, it is possible to achieve much larger displacements
driving motion, it is possible to achieve much larger displacements than with the PZT than with the PZT
actuator alone, while maintaining high precision and accuracy
actuator alone, while maintaining high precision and accuracy levels. For designing high-levels. For designing high-
precisionpositioning
precision positioningCMs, CMs,thethe mostmost common
common configuration
configuration is ofisrigid
of rigid
links links combined
combined with
with flexure
flexure pivots pivots that facilitate
that facilitate bendingbending and around
and twisting twistingcentral
around central
points. points.
Popular Popular
shapes for
shapes
these for these
elastic hingeselastic
includehinges include
circular, circular,
elliptic, elliptic,
parabolic, orparabolic,
[Link] hyperbolic. Of these,
these, semicircular
semicircular
hinges hingesfor
are favored aretheir
favored for theirof
advantages advantages of minor concentrated
minor concentrated stress, immovablestress,rotation
immov-
able rotation center, high accuracy movement transmission,
center, high accuracy movement transmission, and simplicity and ease of fabrication. and simplicity and ease of
fabrication. Therefore, in this design, semicircular elastic hinges
Therefore, in this design, semicircular elastic hinges are used to facilitate the relative are used to facilitate the
relative movement
movement among the [Link]
links. Figure 2 illustrates
2 illustrates the schematic
the schematic designdesign
conceptconcept
of a CMof a
CMa for
for a turning
turning machining
machining application.
application. This This
designdesign combines
combines the lever
the lever and parallelo-
and parallelogram
gram mechanisms
mechanisms to address
to address the design
the design [Link].
In thisIndesign,
this design, the mechanism
the lever lever mechanism is used is
used twice
twice for displacement
for displacement amplification.
amplification. TheThe parallelogram
parallelogram mechanismaims
mechanism aimstotoincrease
increase
rigidity and
rigidity and partially
partially amplify
amplify the the input
input displacement.
displacement. The The coupling
coupling effect
effect among
among these
these
mechanisms, that
mechanisms, that would
would induce
induce unwanted
unwanted motionmotion in in the
the other
other direction,
direction, is is usually
usually thethe
critical
critical concern. The Thecurrent
currentdesign
design intentionally
intentionally usesuses a symmetrical
a symmetrical arrangement
arrangement to
to mit-
mitigate
igate thisthis effect.
effect. Additionally,
Additionally, thetheevaluation
evaluationofofthe theparasitic
parasiticmotion
motion isis also
also imposed as as aa
constraint
constraint function
function in in the
the optimization
optimization designdesign process
process to to carefully
carefully consider
consider thethe coupling
coupling
effect,
effect, which
which can can impact
impactthe theperformance
performanceof ofthe
themechanism.
mechanism.

Figure 2.
Figure 2. Schematic
Schematic of
of aa high-precision
high-precisioncompliant
compliantfeed
feeddrive
drivemechanism.
mechanism.

2.2.
2.2. Displacement
Displacement Amplification
Amplification Analysis
Analysis
Displacement amplification A
Displacement 𝐴amp isisthe
theratio
ratioofofthe
theoutput
output and
and input
input displacement,
displacement,
y𝑦out andandyin
𝑦, respectively. This
, respectively. section
This derives
section derivesthethe
analytical equation
analytical A amp𝐴 using
for for
equation the
using
velocity method, with the translational velocity of the points and the angular
the velocity method, with the translational velocity of the points and the angular velocity velocity of
the links. A simplified dynamic model is used for modeling the CM in Figure
of the links. A simplified dynamic model is used for modeling the CM in Figure 2, where 2, where
all
all links
links are
are assumed
assumed to to be
be rigid,
rigid, and
and revolute
revolute joints
joints replace
replace all semicircular elastic
all semicircular hinges.
elastic hinges.
Due to the symmetrical design of the mechanism, only half of the structure is modeled, as
shown in Figure 3.
 Appl. Sci. 2023, 13, 4968 5 of 29

Appl. Sci. 2023, 13, 4968 5 of 27

Due to the symmetrical design of the mechanism, only half of the structure is modeled, as
shown in Figure 3.

[Link]
Figure Displacementamplification
amplificationanalysis
analysismodel.
model.

ItItisisnoted
notedthat
thatthe
thetranslational
translationalvelocity
velocityof
ofpoints 𝑀, 𝑁,
points M, and P𝑃 can
N, and canbe
beeasily
easilywritten
written
from the angular velocity of the corresponding
from the angular velocity of the corresponding links: links:
VM𝑉= =
ω1𝜔 𝐿
LOM (1)
(1)

VN 𝑉
=ω=2 𝜔
LO1𝐿N = =
ω1𝜔LON
𝐿 (2)
(2)
VP = ω2 LO1 P (3)
𝑉 =𝜔 𝐿 (3)
where, ω1 and ω2 are the angular velocity of the links OM and O1 P, respectively. Therefore:
where, 𝜔 and 𝜔 are the angular velocity of the links 𝑂𝑀 and 𝑂 𝑃, respectively. There-
yout ∼ VP ω2 LO1 P LON LO1 P
fore: A amp = = = = (4)
yin VM ω1 LOM LO1 N LOM
𝑦 method 𝐿
𝑉 to𝜔calculate 𝐿 𝐿
In addition to using the
𝐴 velocity
= ≅ = = the amplification ratio, A amp (4)
can be easily derived using simple 𝑦geometric
𝑉 𝜔 𝐿
relations 𝐿 𝐿the CM has already been
when
approximated
In addition as atorigid mechanism.
using the velocity Some inaccuracies
method maythe
to calculate be amplification
present in thisratio, simplified
𝐴
model, since the internal force, offset of the rotation center of
can be easily derived using simple geometric relations when the CM has already been the flexure hinge, and
deformation of the rigid link are not considered. However, the lengths
approximated as a rigid mechanism. Some inaccuracies may be present in this simplified of all links fulfilling
the specified
model, sincerequirements are promptly
the internal force, offset of determined. This step
the rotation center helps
of the to reduce
flexure hinge, the
andnumber
defor-
of design parameters in the later phase.
mation of the rigid link are not considered. However, the lengths of all links fulfilling the
specified
2.3. requirements
Input/Output Stiffnessare promptly
Analysis determined.
Using the Matrix This step helps to reduce the number of
Method
design parameters in the later phase.
In the general matrix method, each elastic joint has 6 DOFs in its local coordinate, and
is therefore modeled using a full [6 × 6] matrix [32]. In this study, to simplify the model of
2.3. Input/Output Stiffness Analysis Using the Matrix Method
elastic joints, they are assumed to have 3 DOFs, including axial and bending deformation,
In the general
and rotation matrix
around their method,
axis. each elastic
This assumption joint has
is feasible 6 DOFs
because theindesigned
its localmechanism
coordinate,
and is therefore modeled using a full 6 × 6 matrix [32]. In this study,
has a sizeable out-of-plane thickness that can be treated as a plane strain model. The other to simplify the
model of elastic joints, they are assumed to have 3 DOFs, including
three out-of-plane deformations are restricted, and not considered in the calculation model. axial and bending
deformation, and
Consequently, rotation
modeling the around theirwith
elastic joint axis.3 This
DOFsassumption is feasible because
can ensure deformation accuracytheandde-
signed mechanism has a sizeable out-of-plane
allow full description of the design characteristics. thickness that can be treated as a plane
strain
A model. The other
compliance three
matrix [Ci ]out-of-plane
of an elasticdeformations
joint is defined areas
restricted,
a ratio ofand the not considered
displacement
in the calculation  model. TConsequently, modeling the elastic
vector {ui } = u x uy θz when it endures an applied force { Fi } = Fx Fy Mz joint with 3 DOFs can Tensure
[33].
deformation
Follow Hooke’s accuracy
law, thisand{ }
u allow
can full
be description
calculated asof the
[34]: design characteristics.
i

{ui } = [Ci ]{ Fi } (5)

Appl. Sci. 2023, 13, 4968 6 of 27

    
u x  C11 0 0  Fx 
or uy =  0 C22 C23  Fy .
0 C32 C33 Mz
   
θz
The general compliance matrix for different flexure hinge cross-sections has been
proposed and developed by Paros and Weisbord [33,35]. The final form of the matrix for
each hinge should be performed by a coordinate
transformation from its local coordinate
system (Oi ) to a global coordinate system O j [32]. The transformation manipulation of a
compliance matrix from Oi coordinate to O j coordinate can be expressed as follows:
 T
j j j
Ci = Ti Ci Ti (6)
j
The transformation matrix Ti is illustrated in Equation (7):
j j j
Ti = Ri Pi (7)
j j
where, Pi and Ri are the translational matrix and the rotational matrix, respectively, for
j j
the coordinate transformation Oi to O j . The matrices Pi and Ri can be described in
Equations (8) and (9).  
1 0 Py
j
Pi = 0 1 − Px  (8)
0 0 1
 
cosθ −sinθ 0
j
Ri =  sinθ cosθ 0 (9)
0 0 1
where, θ is the rotational angle around the z axis.
In Figure 4a, the flexure hinges Hi (i = 1, 2 . . . 7), as well as the rigid links
Gj ( j = 1, 2 . . . 4), are either serially or parallel connected to form the overall structure
of the mechanism. The calculation for the equivalent stiffness of the hinges or the links
in this section uses the PRBM [36,37]. It is easy to derive the compliance matrices of the
branches ( A), ( B), and (C ) for the left section I:
 T  T  T
COA = TO
H1
C H1
TO
H1
+ T O
G1 C G1
TO
G1 + TO
H2
C H2
TO
H2
(10)
 T  T  T
CO O
B = T H3 C H3 T H3
O
+ TO G2 CG2 TG2
O
+ TO O
H4 C H4 T H4 (11)
 T
CO O
C = TGO CGO TGO
O
(12)
where, C Hi , CGj , TO O
Hi , and TGj are the compliance matrices and transformation matrices of
the flexure hinges Hi and the rigid links Gj from their local coordinate system to the global
coordinate system xOy, respectively.
The compliance matrix for the group ( ABC ) in section I:
 
−1   −1  −1
O O O
C ABC = CC + C A + CO B (13)

The compliance matrices of the branches ( D ), ( E) in section I:

 T  T  T
COD = TO
H5 C H5
TO
H5 + TO
G3 C G3
TO
G3 + TO
H6 C H6
TO
H6 (14)
 T  T
CO O
E = T H7 C H7 T H7
O
+ TO G4 CG4 TG4
O
(15)
The compliance matrix for the group ( DE) in section I:
CO O O
DE = C D + CE (16)
The accumulative compliance matrix for the section I:
CO O O
I = C ABC + C DE (17)
Appl. Sci. 2023,
Appl. 13,2023,
Sci. 496813, 4968 7 of 29 7 of 27

(a)

(b)
FigureFigure
4. Compliant matrix analysis
4. Compliant models:models:
matrix analysis (a) compliant configuration,
(a) compliant and (b) the
configuration, andequivalent
(b) the equivalent
dynamic [Link].
dynamic

The compliance matrixisfor

The mechanism the group
designed (𝐴𝐵𝐶)
with in section I: sections, I and II. Therefore, the
two symmetrical
compliance matrix of the right half section II can be easily determined from CO I : (13)
𝐂 = 𝐂 + (𝐂 ) + (𝐂 )
 T
CO
The compliance matrices of the branches
O
I I = Ry (𝐷),
O
(π )C(𝐸)
I R
O
iny (section
π) I: (18)
The overall
𝐂 output
= 𝐓 𝐂compliance
𝐓 +𝐓 matrix
𝐂 𝐓 of the+whole
𝐓 𝐂 mechanism,
𝐓 with respect
(14) to the
output coordinate system xOy, is calculated as:
𝐂 = 𝐓 O 𝐂 𝐓 O +−𝐓 1 𝐂 𝐓−1 −1
  
O (15)
Cout = C I + CI I (19)
The compliance matrix for the group (𝐷𝐸) in section I:
Figure 4b is the equivalent compliance matrix model of the mechanism that is refer-
𝐂 xO
enced to the input coordinate system = 𝐂in y.+From
𝐂 this model, the input compliance (16)matrix
of the mechanism can be calculated as:
The accumulative compliance matrix for the section I:
  −1   −1  −1     −1
Oin −1 Oin −1
 
O Oin Oin
Cin = C ABC +𝐂 C=Á𝐂 +𝐂 + C DE + C D́ É (17) (20)
B́Ć

The mechanism
Finally, theisinput
designed with two
and output symmetrical
stiffness matricessections, I out
Kin and K andare:
II. Therefore, the
compliance matrix of the right half section II can be −1 determined from 𝐶 :
 easily
Kin = CO in (21)
𝐂 = 𝐑 (𝜋)𝐂  𝐑 (𝜋) (18)
−1
The overall output compliance matrix Kout
of the CO
= whole
out mechanism, with respect to the (22)
output2.4.
coordinate Analysis𝑥𝑂𝑦, is calculated as:
Dynamic system
𝐂 flexure
Using the PRBM [1], the ) + is
= (𝐂 hinge )
(𝐂usually approximated as a traditional
(19) rigid
rotary joint and a torsional spring to simplify and convert the CM into an equivalent
rigid-body model. However, for more general investigation, this paper treats each flexure
 𝐊 = (𝐂 ) (22)

2.4. Dynamic Analysis

Using the PRBM [1], the flexure hinge is usually approximated as a traditional rigid
Appl. Sci. 2023, 13, 4968 rotary joint and a torsional spring to simplify and convert the CM into an equivalent rigid- 8 of 27
body model. However, for more general investigation, this paper treats each flexure hinge
with 3 DOFs exhibiting axial, bending, and torsion deformation, to build the natural fre-
quency
hinge with mathematical model and
3 DOFs exhibiting thebending,
axial, analytical displacement
and control model.
torsion deformation, This
to build dynamic
the natural
model derivation aims to facilitate a quick evaluation of the initial
frequency mathematical model and the analytical displacement control model. This design phase, and dy-
as-
sists in a preliminary prediction for the possibility to control it in the later
namic model derivation aims to facilitate a quick evaluation of the initial design phase, and stage.
assistsIn in
a plane strain model,
a preliminary a flexure
prediction hinge
for the exhibits to
possibility three degree-of-freedoms,
control which in-
it in the later stage.
cludeIntwo translational movements (𝑢 , 𝑢 ) and a rotational movement
a plane strain model, a flexure hinge
exhibits three degree-of-freedoms, which (𝜃 ). The semi-
circular
include two elastic hinge in Figure
translational 5a has been
movements u x , umodelled as a 3-DOF joint in Figure 5b. The
y and a rotational movement ( θz ). The semi-
corresponding spring constants 𝑘 , 𝑘 , and 𝑘
circular elastic hinge in Figure 5a has been modelled as atherepresent three joint
3-DOF displacement
in Figurepossibil-
5b. The
ities.
corresponding spring constants k x , k y , and k z represent the three displacement possibilities.

(a) (b)
Figure 5. Coordinates
Coordinates of a semicircular elastic hinge (a) and its equivalent PRB model (b).

In the mathematical
mathematical model mechanism in Figure 6, the notations ((𝑚
model of the mechanism mi ,, 𝐿Li)),,
(𝑖
( i = 1, 2 … 4),)and 𝜃 (𝑗
2 . . . 4 , and θ j (= 1, 2) are
j = 1, 2 are the mass, length, and rotaion angle
) the mass, length, and rotaion angle of the correspond- of the corre-
sponding
ing rigidrespectively.
rigid links, links, respectively.
The
dynamic The parameters
dynamic parameters
of the model, of including
the model, including
translational
translational
stiffness 𝐾 ,stiffness
𝐾 Kix , Kiy stiffness
and torsional (𝐾 ) stiffness
and torsional (Kiz )hinges,
of the flexure of the flexure
and the hinges,
second mo-and
the second moment of
ment of area of the rigid links area of (𝐼
the), rigid links ( I ) , are presented and calculated
are presented and calculated using the Equations (23)–
i using the
Equations
(26) [33,35,38].(23)–(26) [33,35,38]. 1
Kiz = (23a)
Cθ
∆θz 24R
Cθ = = Γθ (23b)
Mz EhT (2R + T )(4R + T )3
3
r
4R
  q
2 2 2
Γθ = T (4R + T ) 6R + 4RT + T + 6R(2R + T ) T (4R + T )arctan 1 + (23c)
T
1
Kix = (24a)
" Cx r #
∆x 1 2(2R + T ) 4R π
Cx = = p arctan 1 + − (24b)
Fx Eh T (4R + T ) T 2
1
Kiy = (25a)
Cy
∆y 3
Cy = = ( Ψ + Ψ2 + Ψ3 + Ψ4 ) (25b)
Fy 4EhT (2R + T ) 1
Ψ1 = 2R(2 + π ) + πT (25c)
3 2 2


8R 44R + 28RT + 5T
Ψ2 = (25d)
(4R + T )2 T 2
 Appl. Sci. 2023, 13, 4968 9 of 27

p
(2R + T ) T (4R + T ) h i
Ψ3 = q 24R3 T + (π − 80) R4 + 8(3 + 2π ) R2 T 2 + 4(1 + 2π ) T 2 R (25e)
T 5 (4R + T )5
 
8(2R + T )4 −6R2 + 4RT + T2
 r 
4R
Ψ4 = −  q arctan 1 +  (25f)
T
T 5 (4R + T )5
mi L2i
Ii = (26)
12
ppl. Sci. 2023, 13, 4968 where, E is the Young’s modulus of the material; Ti , Ri , and h are the minimum9 of 29 thickness,
radius, and height of the elastic hinges, respectively.

Figure 6. The mathematical model of the CFDM.

Figure 6. The mathematical model of the CFDM.

When the mechanism in Figure 6 is 1 operating, the total dynamic energy is calculated as:
𝐾 = (23a)
4 𝐶2 4 2
1 1 . 2
T0 = ∑ TEi + ∑ REj = ∑ mi v2i + ∑ Ij θ j (27)
∆𝜃 i=1 24𝑅
j =1 i =1
2 j =1
2
𝐶 = = Γ (23b)
Cascading the above𝑀 general (2𝑅 + 𝑇)(4𝑅
𝐸ℎ𝑇 formula + 𝑇)
for each link, Equation (28) can be easily obtained:
.
2 .
2 .2 .
2 .2 .
2
T0 = 12 m1 yin + m2 y2c + I2c θ 1 + 2m3 y3c + 2I3c θ 2 + 12 m4 y5
.
2
 . . 2 . .   . . 2
y2 +y3 + 4𝑅𝑇 +y3𝑇−y)2 + 26𝑅(2𝑅 + y𝑇)
 . . 2 4𝑅 (28)
= 12 m1Γ y= 𝑇(4𝑅
+ m+2 𝑇)(6𝑅 + I2c L1 + L2 + 2m3 4 2 5 𝑇(4𝑅
+y +3c𝑇)arctan
y5 − y4 1++1 m4 y. 2 (23c)


in 2 + 2I L3 + L4 2 𝑇 5

The total strain energy is also calculated using Equation (29):

4 2 1
V = ∑ 12 Kiy ∆y2i𝐾+ = ∑ 𝐶12 K jz θ 2j (24a)
i =1 j =1
2
= K1y (y1 − yin ) + K2z θ1 + K3z θ2 + K2x y2 + K3y (y2c − y3 )2 + 2K4x y24 + 2K5x (y3c − y5 )2
+ K1z θ12 2 2 2
 ∆𝑥 1 2(2𝑅
2 + 𝑇)y3 −y2 2 4𝑅 𝜋2 (29)
y3 − y3 − y2
= K1y y2 + 𝐶L1L(= y2 )
= − y in + K 1z arctan+ 1
K +
2z −2 (24b)
2 1 𝐹2 𝐸ℎ 𝑇(4𝑅 + 𝑇L1 + L2
+L L +L
 2𝑇)1 2 2
+K3z Ly53 − y4
+ L4 + K2x y22 + K3y y2 −2 y3 + 2K4x y24 + K5y y4 −2 y5
1
 . 
where, y j , y j are the displacement 𝐾 = and velocity of the links, respectively.(25a)
Their relations
𝐶 (30a) and (30b).
are illustrated in the system of Equations

∆𝑦 3
𝐶 = = (Ψ + Ψ + Ψ + Ψ ) (25b)
𝐹 4𝐸ℎ𝑇(2𝑅 + 𝑇)
Appl. Sci. 2023, 13, 4968 10 of 27



 y1 = y2 + L1 θ1
 (L +L )



 y2c = y2 + 1 2 2 θ1
y3 = y2 + ( L1 + L2 ) θ1

(30a)

 y3 = y4 + L3 θ2
 (L +L )
y = y4 + 3 2 4 θ2

 3c



y = y + ( L + L )θ
 5 . 4 . 3 .4 2
y =y +L θ
 . 1 . 2 ( L1 +1 L21) .


 .y2c =. y2 +


 2 θ1
 .
y3 = y2 + ( L1 + L2 ) θ 1

. . . (30b)
 y3 = y4 + L3 θ 2
.

. .

 ( L3 + L4 )
 y = y + θ 2
 . 3c . 4


 2 .
y5 = y4 + ( L3 + L4 ) θ 2
Using the Lagrangian method with Γ = T0 − V,
 
d ∂Γ ∂Γ
. − = Qi i = 1, 2, . . . , 4 (31)
dt ∂yi ∂yi
Replacing Equations (28)–(30) into Equation (31), the general movement equation of
the mechanism is:
..
My + Ky = 0 (32)
where, M and K are the mass and stiffness matrices, respectively.
The elements of the matrices M and K are determined as follows:
∂2 T0
mij = . . ; i, j = 1, 2, . . . , n (33a)
∂ yi ∂ y j
∂2 V
k ij = ; i, j = 1, 2, . . . , n (33b)
∂yi ∂y j
The natural frequency ω0 of the mechanism can be easily found by solving the charac-
terstic determinant in Equation (34) [39]:
 
Det K − ω 2 M = 0 (34)

3. Optimization Design
In this design process, a universal two-step design algorithm is proposed, in which
PRBM optimization and reliability-based multi-objective NSGA-II algorithm optimization
are combined, as shown in Figure 7. All of these combinations form an RBDO for the CM
in this section. The two implementation steps of the algorithm are specified as follows:
1. PRBM optimization The problem and the DVs are defined and formulated by estab-
lishing the mathematical relation between DVs and the objective function. This is
then solved using the “fmincon” function in MATLAB.
2. Reliability analysis integrated multi-objective NSGA-II optimization for the CM As a
rule of thumb, positioning CM optimization design usually involves a displacement
amplification target. One can include other cost functions depending on the specific
applications of the mechanism. To ensure the micromachining capability of the CFDM,
it must make adequate output displacements in the cutting direction, and be available
for high-speed machining. A tool exhibiting a high natural frequency can facilitate
high-speed cutting. Therefore, in this optimization step, the two selected objective
functions are: large displacement amplification ratio and high natural frequency. This
paper implements the multi-objective design optimization problem using the NSGA-II
algorithm in MATLAB.
[Link].
Appl. Sci.2023,
2023,13,
13,4968
4968 12 29
11 of 27

Figure7.7. Reliability-based
Figure Reliability-baseddesign
designoptimization
optimizationusing
usingNSGA-II.
NSGA-II.

Instead
Insteadofofonly
onlysolving
solvingaa common
common multi-objective
multi-objective design
design optimization
optimization problem
problemwith
with
deterministic
deterministic constraints,
constraints, this
this paper
paper integrates
integrates the
the reliability
reliability analysis
analysis into
into the
the NSGA-II
NSGA-II
optimization
optimization process,
process, which
which imposes
imposes thethe probability
probability ofof violating
violating the
theoriginal
originalconditions
conditions
to
to form an RBDO problem. During the optimization process, the algorithm evaluates the
form an RBDO problem. During the optimization process, the algorithm evaluates the
reliability
reliabilityof
ofeach
eachdesign
designininevery
everyevolution
evolution generation
generation using
using the
the FORM.
FORM. The The uncertainty
uncertainty
of
ofthe
theinput
input parameters,
parameters, such
such asas cutting
cutting forces
forces due
due to
to the
the inhomogeneity
inhomogeneity of of the
the materials,
materials,
the
thedeviated
deviateddimensions
dimensionsofofthe
the fabricated
fabricated CFDM
CFDM prototype,
prototype,and and undesirable
undesirable mismatching
mismatching
during
duringthetheassembly,
assembly,are
areconsidered
consideredand andformulated
formulatedasaslimit
limitstate
statefunctions.
functions.

3.1.
3.1. Pseudo-Rigid-Body
Pseudo-Rigid-Body Model
Model Optimization
Optimization
In
In this section, the optimizationproblem
this section, the optimization problemisisformulated
formulatedusing
usingthe
thePRBM
PRBMofofthethemecha-
mech-
nism in Figure 3, and the displacement amplification ratio in Equation (4). Other precision-
anism in Figure 3, and the displacement amplification ratio in Equation (4). Other preci-
positioning CMs often choose displacement amplification maximization as an objective.
sion-positioning CMs often choose displacement amplification maximization as an objec-
However, the trade-off would be a worse positioning resolution and a nonlinear in-
tive. However, the trade-off would be a worse positioning resolution and a nonlinear in-
put/output displacement relation if A amp is excessively large. In this design, an adequate
put/output displacement relation if 𝐴 is excessively large. In this design, an adequate
displacement gain of A amp = 3 is chosen as a compromise, considering the limited maxi-
displacement gain of 𝐴 = 3 is chosen as a compromise, considering the limited max-
mum stroke of the PZT actuator and its unexpectedly increased resolution. The DVs and
imum stroke of the PZT actuator and its unexpectedly increased resolution. The DVs and
their limits, the objective function, and the necessary constraint functions are summarized,
their
as limits,
shown in the objective
Table function,
1. In the and the
optimization necessary
problem constraint
of the functions in
PRBM structure arethis
summarized,
study, the
as shown in Table 1. In the optimization problem of the PRBM structure in this study, the
Appl. Sci. 2023, 13, 4968 12 of 27

“fmincon” function in MATLAB is used to optimize Equation (35). Table 2 presents the
optimal results for the DVs.

Table 1. PRBM optimization formulation.

1. Objective function: Min A amp − 3 (35)

2. DVs: LOM , LON , LO1 N , LO1 P
3. Subjected to:
(i) Limits of the DVs (mm)
10 ≤ LOM ≤ 20 (36)
15 ≤ LON ≤ 30 (37)
10 ≤ LO1 N ≤ 20 (38)
25 ≤ LO1 P ≤ 40 (39)
(ii) Boundary constraints:
LON + LO1 N = 40 (40)
LO1 P − LO1 N ≥ 20 (41)

Table 2. The optimal results for the DVs.

DVs Value (mm) DVs Value (mm)

LOM 16.5 LO1 N 17
LON 23 LO1 P 37

For the simple mechanism configuration in Figure 2, PRBM optimization can be

unnecessary. One can also quickly and straightforwardly solve it using the provided
system of equations. However, this paper proposes a general optimization procedure that
could work for more complicated CMs. As mentioned in Section 2.2, the PRBM aims to
determine the length of rigid links, quickly fulfilling the preliminary requirements. This
optimization step helps reduce the number of design parameters in the next section.

3.2. Reliability-Based Multi-Objective NSGA-II Optimization

Figure 4 shows the CM configuration corresponding to the PRB model in Figure 3.
The designated design space of the mechanism is limited to (100 × 50 × 30) mm3 , which is
adequate for the later task of assembly into an available machine tool. The previous PRBM
optimization algorithm has constrained this design space in its problem formulation. There-
fore, it can be easily changed in appropriately adapting to any other design configurations.
The general working principle of the CM is to rely on the deformation of the mechanism
to transfer or transmit motion. As a rule of thumb, this deformation must be maintained
within the elastic limit of the fabricated material, and prolonging of its fatigue endurance
is allowed. Generally,

the maximum concentrated stress (σmax ) must be lower than the
yield strength σy when the structure experiences the most significant displacement. For
hinge-type CMs, σmax occurs when a semicircular elastic hinge reaches its largest rotation
angle θmax [40,41], namely:
9
E(1 + η) 20 σy
σmax = 2 θmax < (42)
η f (η) SF
where, η = T/2R is a dimensionless coefficient; T and R are the minimum thickness and
radius of a hinge, respectively; f (η) is a function of ∆ [40,41], and SF is the safety factor.
Equation (5) in Section 2.3 is used in evaluating the displacement vector, including the
rotation angle θz of a hinge if given the tensile force Fx , transverse force Fy , and bending
moment Mz . This calculation does not consider the effect of out-of-plane loads, since the
mechanism has constrained its deformation in the xOy plane by designing a much larger
out-of-plane thickness h than the minimum in-plane thickness T of each hinge. The more
flexible the CM, the more easily it could deform. Reducing the thickness T of the elastic
Appl. Sci. 2023, 13, 4968 14 of 29

Appl. Sci. 2023, 13, 4968 13 of 27

the elastic hinge is one of the common alternatives. Therefore, the DVs at this calculation
step are
hinge the minimum
is one of the common thickness of the hinges
alternatives. (𝑇 , 𝑇the
Therefore, , 𝑇DVs
), asatshown in Figure step
this calculation 2. are the
Before conducting the size optimization using
minimum thickness of the hinges (T1 , T2 , T3 ), as shown in Figure 2. the multi-objective NSGA-II optimiza-
tion, Before
in this conducting
section, thethe sensitivity of the DVs is analyzed to understand
size optimization using the multi-objective NSGA-II optimiza- the influence of
each variable. The knowledge gained from this could allow eliminating
tion, in this section, the sensitivity of the DVs is analyzed to understand the influence parameters that
have
of eachlittle effect, and
variable. Theexpanding
knowledgethe boundary
gained fromof those
this couldthatallow
haveeliminating
more impactparameters
on the de-
siredhave
that objective functions.
little effect, and In addition,the
expanding theboundary
convergence speed
of those of have
that the objective
more impactfunctions
on the is
enhanced, which reduces the consumption of computer resources
desired objective functions. In addition, the convergence speed of the objective functions and processing time.
Figure
is 2 shows
enhanced, all of
which the design
reduces parameters for
the consumption of the CM model,
computer resourcesincluding the dimensions
and processing time.
of the hinges (𝑇 𝑎𝑛𝑑 𝑅 ) and the length of the rigid links (𝐿
Figure 2 shows all of the design parameters for the CM model, including theoptimization ). The second dimensions
step
of thefacilitates
hinges ( Tobservation of the output responses of the mechanism, including input
i and Ri ) and the length of the rigid links ( Li ). The second optimization
stiffness,
step natural
facilitates frequency, of
observation displacement
the output amplification,
responses of the andmechanism,
stress, whichincluding
are sensitive
inputto
any variation
stiffness, in the
natural DVs. Figure
frequency, 8 illustrates
displacement the sensitivity
amplification, analysis
and stress, results
which arefor the DVs.
sensitive to
The variation
any analysis shows
in the that
[Link] input8stiffness
Figure illustrates of the
the sensitivity
mechanismanalysis increases proportionally
results for the DVs. to
the thickness, as seen in Figure 8a. Since the natural frequency
The analysis shows that the input stiffness of the mechanism increases proportionally to in Figure 8b is a function
of the
the input stiffness,
thickness, as seen in it also
Figure increases
8a. Sinceaccordingly.
the natural Figure
frequency 8c ofinthe displacement
Figure amplifi-
8b is a function of
cation
the input confirms,
stiffness,again,
it alsothat the larger
increases the stiffness,
accordingly. Figure the8cmore
of thecomplex the mechanism
displacement amplification de-
[Link],
confirms, Apartthat fromthe the apparent
larger effect the
the stiffness, of 𝑇more and 𝑇 , thethe
complex impact of 𝑇 shows
mechanism a con-
deformation.
Apart from the since
trary forecast, apparent it supports T2 anddisplacement
effect of slight T3 , the impactamplification.
of T1 shows a In contrary forecast,
this figure, it issince
also
it supportsthat
apparent slight 𝐴
thedisplacement amplification.
is heavily affected byInthe thisthickness
figure, it ofis also apparent
the hinges, A amp
that theEqua-
whereas
is heavily
tion (4) doesaffected by thethe
not involve thickness
parameter 𝑇. hinges,
of the This secondwhereas Equationstep
optimization (4) does not involve
will compensate
the parameter T. This second optimization step will compensate
for any disadvantages of the previous model. Finally, the sensitivity of the DVs in Figure for any disadvantages of
the previousinversely
8d appears model. Finally, the sensitivity
and nonlinearly of the DVs
proportional in Figure
to the 8d appears
concentrated stressinversely
value to and
dif-
nonlinearly
fering degrees. proportional to the concentrated stress value to differing degrees.

Figure 8. DV sensitivity analysis to the objective functions: (a) input stiffness;

Figure stiffness; (b) natural frequency;
(c) amplification
(c) amplification ratio;
ratio; (d)
(d) maximum
maximum stress.
stress.
Appl. Sci. 2023, 13, 4968 14 of 27

Table 3 describes the boundary of the DVs, constraint, and objective functions of this
problem. The objective function f 1 in Equation (43) aims to maximize the first natural
frequency to prevent any resonant phenomena during the operation of the mechanism.
The second objective function f 2 in Equation (44) aims to maximize the displacement
amplification ratio A amp . At first glance, it looks contradictory to the optimization target in
the previous section in Equation (35), in which the aim is to maintain A amp close to a given
value of 3.0. Maximizing the function f 2 ensures that it should not fall below 3.0. Due to
some simplification assumptions in the PRBM, when more authentic parameters of elastic
structures are considered, it is hard to maintain the A amp at the expected ratio. To eliminate
infeasible designs during the process of solving the multi-objective design optimization
problem, the candidates must fulfill the constraints of stress concentration requirements for
the elastic hinges in Equation (52).

Table 3. Multi-objective design optimization formulation.

1. Objective functions:
f 1 = Max( f 0 ) (43)
yout (44)
f 2 = Max yin

2. Limit state function (*):

{ G ( Fi , u x , σm ) − D0 ≤ 0} ≥ Φ(β), i = x, y, z (45)
Reliability R0 ≥ 98%
3. DVs: T1 , T2 , T3
4. Subjected to:
(i) Boundary of the DVs (mm)
0.4 ≤ T1 , T2 , T3 ≤ 1.5 (46)
(ii) Input stiffness requirement:

Kin ≤ 0.15K pzt K pzt = 480 N/µm (47)
(iii) Force requirement (N) (*):
g1 = 50 − Fz−cutting f orce ≤ 0 (48)
Fz−cutting f orce ∼ Normal (50, 10)
g2 = 50 − Fx− f eeding f orce ≤ 0 (49)
Fx− f eeding f orce ∼ Normal (50, 10)
g3 = 200 − Fy−radial f orce (50)
Fy−radial f orce ∼ Normal (200, 50)
(iv) Parasitic motion in x − axis (µm) (*):
g4 = 0.01 − u x ≤ 0 (51)
u x ∼ Normal (0.01, 0.005)
(v) Concentrated stress:
σ
g5 = σm ≤ SFy (52)

With the reliability analysis step integrated into the NSGA-II optimization process,
constraint equations with an asterisk are supplemented. In this application, to assure that
the structure has micron resolution, adequate rigidity, and allowable displacement error,
the parasitic motion in the x-direction in Equation (51) must be satisfied, and treated as
a limit state function. In addition, to ensure that the CFDM is machineable, limit state
functions of cutting force are established, satisfying the Equations (48)–(50). It is chosen for
this analysis task since the current objectives are implicit functions, and given the relative
advantages of the FORM in dealing with such problems compared with the Monte Carlo
and SORM methods, as analyzed in Section 1.
The NSGA-II optimization in this step is implemented in MATLAB using the AN-
SYS ADPL as the medium to convey the calculation process. The plane strain element
(PLANE82) is used in the analysis to evaluate the constraint functions (stiffness, maxi-
mum stress, parasitic motion) and objective functions (amplification ratio and natural
frequency). Although the algorithm can directly use the finite element method (FEM) at
Appl.
Appl. Sci. 2023, 13, 4968 15of
16 of 27
29

the beginning
beginning for global
for global computation,
computation, this research
this research still suggests
still suggests usingusing this engineering
this engineering com-
computational tool after
putational tool after the theoretical
the theoretical PRB model.
PRB model. This analysis
This analysis sequencesequence is beneficial
is beneficial to thor-
to thoroughly
oughly understand
understand a new before
a new design designone before
can one can relinquish
relinquish the entirethe entiretoprocess
process commer- to
commercial engineering
cial engineering [Link].
The material used to fabricate
fabricate the
the mechanism
mechanism is is aluminum
aluminum alloy
alloy (Al
(Al 7075–T6).
7075–T6). Its
mechanical properties are Young’s modulus E 𝐸 71.7 GPa, Poisson’s ratio 𝜗 = 0.33, yield
= 71.7 GPa, Poisson’s ratio ϑ
3
strength σ𝜎y = = 503
503MPa,
MPa, and
and density ρ𝜌= 2810 kg⁄m . .The
= 2810kg/m Thetotal
totalnumber
number of of plane elements
in the FEM
in the FEM model
modelisis22,638
22,638elements.
elements. In the
In the RBDO-NSGA-II
RBDO-NSGA-II optimization
optimization process
process in
in Fig-
Figure 7, the number of generations is preliminarily set as 100 and the population
ure 7, the number of generations is preliminarily set as 100 and the population as 20. The as 20.
The optimization
optimization process
process cancan
alsoalso
be be terminated
terminated bybyanother
anotherconvergence
convergencecriteria
criteria when
when thethe
population has no further evolution. Figure 9 shows a Pareto front of this
population has no further evolution. Figure 9 shows a Pareto front of this RBDO-NSGA- RBDO-NSGA-II
optimization
II optimization (Pareto P2)P2)
(Pareto with
withtwo objective
two objectivefunctions.
functions.

Figure 9.
Figure 9. Pareto
Pareto diagrams of DO
diagrams of DO and
and RBDO
RBDO solutions.
solutions.

research, another
In this research, another determined
determined optimization (DO), without considering any
provided limit limit state
statefunctions,
functions,isis also implemented
also implemented to evaluate
to evaluatethe reliability analysis
the reliability con-
analysis
tribution in the
contribution RBDO-NSGA-II
in the RBDO-NSGA-II optimization.
optimization. ThisThisoptimization
optimizationresult (Pareto
result P1) isP1)
(Pareto also
is
presented
also presented in Figure
in Figure9 for9comparison
for comparison withwith
its counterpart.
its counterpart.
To
To further
further elaborate
elaborate on the the difference
difference between
between the DO and the RBDO results, this
section will use use several
severalsolutions
solutionslying
lyingon onthe
thetwo
two Pareto
Pareto fronts
fronts (P1(P1
andandP2)P2) in Figure
in Figure 9 for9
for comparison.
comparison. Table Table 4 presents
4 presents the benchmarking
the benchmarking results,
results, where where
pointspoints
(C, E)(C,
areE)two aresolu-
two
solutions of Pareto
tions of Pareto P2, and P2, and points
points (A,D)B,are
(A, B, D) three
are three solutions
solutions of Pareto
of Pareto P1. P1.
In Table 4, the reliability index β is the ratio of the mean of a random variable to its
standard [Link]
Table 4. Comparison DOpurpose
and RBDO of solutions.
this table is to illustrate that β significantly influences
the optimal results. The FORM analysis evaluates β in a calculation loop. Generally,
DO-NSGA-II RBDO-NSGA-II
Parameters after
Point A three iterations, Point Bβ is convergedPoint and Dwill be used toPoint accessC the reliability level
Point E ( R0 ).
𝑇 (mm) For instance, at the1.2extreme of the amplification
0.48 1.4 and natural
1.2 frequency objectives, 1.15 the
𝑇 (mm) optimal results of 0.46
0.84 the RBDO problem 0.5 at point C (3.104; 1500) 0.61 are smaller than 0.61those of
𝑇 (mm) the DO problem at0.63
0.5 point A (3.405; 1500). 0.74 However, when0.6 using the FORM assessment0.61
Maximum stress (MPa) 84.76
method to calculate 73.56the reliability index 77.33at point A, the β73.7 values for ( β 1 , β 2 ,72.33
β 3 , and β 4 )
Input displacement (μm) 16.25 13.81 13.63 13.5 13.5
are (2.25, 10.88, 0.0, and 0.0), respectively. The safety level of a solution is evaluated via
Input stiffness (𝐾 ) 48.21 53.11 53.03 59.26 59.26
Natural frequency 𝑓 (Hz)
the
1500 3
β index. This value
1499
of β 3 is R =
0 1499 42.91% for point A.
1500
At point C, the β values
1499
for
Amplification ratio 𝑓 ( β , β
3.4051 2 3 , β , and β ) are
4 3.13 ( 4.11, 14.20, 3.0,
3.105and 3.0 ) , respectively.
3.104 The β 3 value for
3.101 point C
𝛽 2.25 has a reliability
(98.77%) 4.20 level at R0 = 99.87%,
(100%) 8.92 which is much 4.11
(100%) higher than that of 8.89
(99.99%) point A. (100%)
Another
𝛽
𝛽 10.88 (100%) 12.96 (100%) 11.58 (100%) 14.20 (100%) 9.49 (100%)
Appl. Sci. 2023, 13, 4968 16 of 27

comparison between point E (3.101; 1499) and point B (3.13; 1499), with approximate
values of objective functions, also shows that the level of safety at point E is R0 = 99.87%;
meanwhile, point B only has a safety level at R0 = 48.26%.

Table 4. Comparison of DO and RBDO solutions.

DO-NSGA-II RBDO-NSGA-II
Parameters
Point A Point B Point D Point C Point E
T1 (mm) 0.48 1.2 1.4 1.2 1.15
T2 (mm) 0.84 0.46 0.5 0.61 0.61
T3 (mm) 0.5 0.63 0.74 0.6 0.61
Maximum stress (MPa) 84.76 73.56 77.33 73.7 72.33
Input displacement (µm) 16.25 13.81 13.63 13.5 13.5
Input stiffness (Kin ) 48.21 53.11 53.03 59.26 59.26
Natural frequency f 1 (Hz) 1500 1499 1499 1500 1499
Amplification ratio f 2 3.405 3.13 3.105 3.104 3.101
β1 2.25 (98.77%) 4.20 (100%) 8.92 (100%) 4.11 (99.99%) 8.89 (100%)
β2 10.88 (100%) 12.96 (100%) 11.58 (100%) 14.20 (100%) 9.49 (100%)
β FORM
β3 0.00 (42.91%) 0.00 (48.26%) 1.77 (96.13%) 3.00 (99.87%) 3.05 (99.88%)
β4 0.00 (43.11%) 0.004 (50.16%) 1.30 (90.24%) 3.00 (99.87%) 3.10 (99.90%)

As seen in Figure 9, the Pareto P1 has an outlier of point A, whose displacement

amplification is larger than the remaining designs of the last DO-NSGA-II population. It
is clear that due to the smaller thickness of T1 , its stiffness is decreased, which makes the
mechanism easier to deform. In the meantime, the Pareto P2 curve is partially coincident
and partly upward with the Pareto P1 curve. It has fewer solutions than the Pareto P1 curve,
since some offspring of their last generation were converged and coincident. There is no
outlier in the Pareto P2, indicating that the reliability index significantly affects the optimal
results. In the set of Pareto P1 optimal solutions, some solutions satisfy the expected value
of R0 > 98%, but some do not. For the reliability concern, all of the optimal solutions
of P2 meet the given R0 value. In a nutshell, it is possible to confirm the praiseworthy
compensation of the RBDO compared to without it.
The target of the RBDO is to simultaneously maximize the natural frequency f 1 and the
displacement amplification f 2 . As seen in the reliability-based Pareto diagram in Figure 9,
there are many design points with different levels of reliability. Among these alternatives,
the chosen optimal design (point C) with the reliability index β 3 = 3, the safety level
R0 = 99.87%, f 1 = 1500 Hz, and f 2 = 3.104, fulfills the given problem with the required
constraints, objective functions, and expected reliability. Table 5 reports these results.

Table 5. Reliability analysis optimization results (point C).

Design Variables Values (mm) Design Variables Values

T1 1.2 f2 3.104
T2 0.61 σmax
73.7 MPa
T3 0.6 Dout Dy 40.0 µm
f1 1500 Hz δx 0 µm
R0 99.87% β3 3

3.3. Evaluation of Results and Discussion

This step aims to verify the stress, displacement, and natural frequency results for
optimal design. The optimal design found in Tables 2 and 5 was reconstructed in a 3D
model using INVENTOR 2018 software, and simulated on ANSYS 18.2 Workbench software.
Figures 10 and 11 illustrate the simulation results.
 Appl.
Appl. Sci. Sci.
2023, 13,2023,
496813, 4968 18 of 29
17 of 27

(a) (b)

(c)

FigureFigure
10. [Link]
FEM simulation
results:results: (a)displacement,
(a) total total displacement, (b) displacement
(b) displacement in the x-direction,
in the x-direction, and and
(c) equivalent stress.
(c) equivalent stress.

As shown in Figure 10, with a given input displacement yin = 13.5 µm provided by
the PZT actuator, the output displacement yout = 39.65 µm in the y-direction is induced,
which is equivalent to the amplification ratio A amp = 2.937, and the parasitic motion in
the x-direction of the tool fixture is completely eliminated (δx = 0 µm). The maximum
concentrated stress σmax = 73.705 MPa satisfies the constraint for the allowable strength of
the material. The finite element method ANSYS ADPL is used to evaluate the stiffness of
the structure, including the input stiffness (Kin ) and output stiffness (Kout ). To calculate
Kin , an input force Fin = 800 N is applied to the input port, and the displacement of this
Appl. Sci. 2023, 13, 4968 18 of 27

point is measured as yin = 13.5 µm. It is easy to calculate that Kin = 59.26 N/µm. The
output stiffness is also calculated in the same way, by applying the output force Fout = 50 N
to the output port, and the obtained displacement is yout = 39.65 µm. Therefore, the output
Appl. Sci. 2023, 13, 4968 stiffness Kout = 5.2 N/µm. The relation between the input and output displacement 19 of is
29
illustrated in Figure 11, which is stably linear in the operational range from 0 to 13.5 µm.

Figure 11. The simulated results of input displacement vs. output displacement and stress.

As shown in Figure 10, with a given input displacement 𝑦 = 13.5 μm provided by

the PZT actuator, the output displacement 𝑦 = 39.65 μm in the y-direction is induced,
which is equivalent to the amplification ratio 𝐴 = 2.937, and the parasitic motion in
the x-direction of the tool fixture is completely eliminated (𝛿 = 0 μm). The maximum
concentrated stress 𝜎 = 73.705 MPa satisfies the constraint for the allowable strength
of the material. The finite element method ANSYS ADPL is used to evaluate the stiffness
of the structure, including the input stiffness (𝐾 ) and output stiffness (𝐾 ). To calcu-
late 𝐾 , an input force 𝐹 = 800 N is applied to the input port, and the displacement of
this point is measured as 𝑦 = 13.5 μm. It is easy to calculate that 𝐾 = 59.26 N⁄μm. The
output stiffness is also calculated in the same way, by applying the output force 𝐹 =
50 N to the output port, and the obtained displacement is 𝑦 = 39.65 μm. Therefore, the
output stiffness 𝐾 = 5.2 N⁄μm. The relation between the input and output displace-
ment is illustrated in Figure
Figure
Figure 11. The
11. The 11, which is
simulated
simulated
stably
results
results ofoflinear in the operational
inputdisplacement
input displacement range
vs. output
vs. output from 0 toand
displacement
displacement andstress.
stress.
13.5 μm.
In the numerical In evaluation
Astheshown of Figure
numerical
in the first natural
evaluation
10, with aof frequency
the first
given in
input Figure
natural 12a, it can
frequency
displacement 𝑦 be
in seen
= Figure
13.5 μm12a, it can be
provided by
that 𝑓 = 1500 Hz. There
seen
the thatis
PZT f 0a =
slight
actuator, 1500 discrepancy
theHz. There
output between
is a slightthe
displacement 𝑦FEM=simulation
discrepancy39.65 μm in
betweenand the
the
the op- simulation
FEM
y-direction and
is induced,
timal results in Table
the 4, due
optimal
which toresults
the round-up
is equivalent the of
intoTable 4,the design
due
amplification parameters
to the 𝐴
round-up
ratio for
= better
of2.937, fabrication
the design
and theparameters
parasitic for better
motion in
in the following step.
the x-direction of the tool fixture is completely eliminated (𝛿 = 0 μm). The maximum
fabrication in the following step.
concentrated stress 𝜎 = 73.705 MPa satisfies the constraint for the allowable strength
of the material. The finite element method ANSYS ADPL is used to evaluate the stiffness
of the structure, including the input stiffness (𝐾 ) and output stiffness (𝐾 ). To calcu-
late 𝐾 , an input force 𝐹 = 800 N is applied to the input port, and the displacement of
this point is measured as 𝑦 = 13.5 μm. It is easy to calculate that 𝐾 = 59.26 N⁄μm. The
output stiffness is also calculated in the same way, by applying the output force 𝐹 =
50 N to the output port, and the obtained displacement is 𝑦 = 39.65 μm. Therefore, the
output stiffness 𝐾 = 5.2 N⁄μm. The relation between the input and output displace-
ment is illustrated in Figure 11, which is stably linear in the operational range from 0 to
13.5 μm.
In the numerical evaluation of the first natural frequency in Figure 12a, it can be seen
that 𝑓 = 1500 Hz. There is a slight discrepancy between the FEM simulation and the op-
timal results in Table 4, due to the round-up of the design parameters for better fabrication
in the following step.

(a)

Figure 12. Cont.

(a)
 Appl. Sci. 2023, 13, 4968 20 of 29
023, 13, 4968
Appl. Sci. 2023, 13, 4968 20 of 29 19 of 27

(b) (b)
Figure
Figure 12. Evaluation 12. Evaluation
of the of the (a)
natural frequency: (a) simulation result; (b) experimental result.
Figure [Link] frequency:
Evaluation simulation
of the natural result;
frequency: (a)(b) experimental
simulation result;result.
(b) experimental result.

4. Experiment 4.
4. Experiment
Experiment
The
Theoptimum
optimum design
designofof
thethe
CFDM
CFDM in Section 3 is 3fabricated
in Section by wire-EDM
is fabricated cutting
by wire-EDM from
cutting
The optimum design of the CFDM in Section 3 is fabricated by wire-EDM cutting
Al 7075–T6,
from Al and is and
7075–T6, shownis in Figure
shown in 13. Figure
Figure 13. 14 shows
Figure 14 the experimental
shows the setup tosetup
experimental test the
to
from Al 7075–T6, and is shown in Figure 13. Figure 14 shows the experimental setup to
prototype to assess its performance.
test the prototypetest the prototype
to assess to assess its performance.
its performance.

FigureCFDM
Figure
Figure 13. The fabricated 13. The
13. Theprototype.
fabricated CFDM
fabricated CFDM prototype.
prototype.

In this experiment, a piezoelectric actuator P-225.10 (Physik Instrumente PI, Karlsruhe,

Germany) drives the CFDM. This PZT actuator can provide a motion of up to 15 µm,
and the stiffness in the motion direction is 480 N/µm. It is suggested that the stiffness of
the piezo actuator should be larger than the inner stiffness of the mechanism to ensure
stable and improved operation. The optimization process in Section 3.2 has imposed
this constraint in its inequality (46). A piezo controller E-470.20 (Piezo System, PI, Inc.,
Karlsruhe, Germany) amplifies the input voltage from a function generator. Two laser
displacement sensors, LK-G30 (Keyence Corporation, Osaka, Japan), with a measuring
range of ±5 mm, a resolution of 0.05 µm, and a maximum sampling frequency up to
50 kHz, are used to measure the input, output, and parasitic motion. The experiment
uses two microtranslation stages to calibrate the focus of the laser sensors. An NI myRIO
1900 control card (National Instruments, Austin, TX, USA) generates an analog signal,
and stores the reading from the displacement sensors with the sampling rate of 50 kHz.
LabVIEW 2014 software performs the deterministic real-time control of the CFDM system.
These devices are mounted on an antivibration table. It isolates external vibrations that
Appl. Sci. 2023, 13, 4968 20 of 27

could affect the accuracy of the measurement results. Figure 15 shows an actual experimen-
tal layout. For practical turning machining, the effect of the cutting force on the mechanism
is dynamic and very significant. However, an ideal testing condition is assumed, to verify
Appl. Sci. 2023, 13, 4968 the theoretical and simulation model in this experimental setup. The RBDO problem has
21 of 29
considered all variations in an authentic manufacturing environment.

Figure 14.
Figure 14. Experimental
Experimentalsetup
setupforforopen-
open-and closed-loop
and control.
closed-loop (1) Driver
control. E-470.20;
(1) Driver (2) PZT;
E-470.20; (2) (3)
PZT;
compliant mechanism; (4) laser sensor LK-G30; (5) microtranslation stage; (6) controller LK-3001P;
(3) compliant mechanism; (4) laser sensor LK-G30; (5) microtranslation stage; (6) controller LK-3001P;
Appl. Sci. 2023, 13, 4968 (7) computer; (8) NI MyRio 1900. 22 of 29
(7) computer; (8) NI MyRio 1900.

In this experiment, a piezoelectric actuator P-225.10 (Physik Instrumente PI, Karls-

ruhe, Germany) drives the CFDM. This PZT actuator can provide a motion of up to 15 μm,
and the stiffness in the motion direction is 480 N⁄μm. It is suggested that the stiffness of
the piezo actuator should be larger than the inner stiffness of the mechanism to ensure
stable and improved operation. The optimization process in Section 3.2 has imposed this
constraint in its inequality (46). A piezo controller E-470.20 (Piezo System, PI, Inc., Karls-
ruhe, Germany) amplifies the input voltage from a function generator. Two laser displace-
ment sensors, LK-G30 (Keyence Corporation, Osaka, Japan), with a measuring range of ±5
mm, a resolution of 0.05 μm, and a maximum sampling frequency up to 50 kHz, are used
to measure the input, output, and parasitic motion. The experiment uses two microtrans-
lation stages to calibrate the focus of the laser sensors. An NI myRIO 1900 control card
(National Instruments, Austin, TX, USA) generates an analog signal, and stores the read-
ing from the displacement sensors with the sampling rate of 50 kHz. LabVIEW 2014 soft-
ware performs the deterministic real-time control of the CFDM system. These devices are
mounted on an antivibration table. It isolates external vibrations that could affect the ac-
curacy of the measurement results. Figure 15 shows an actual experimental layout. For
practical turning machining, the effect of the cutting force on the mechanism is dynamic
and very significant. However, an ideal testing condition is assumed, to verify the theo-
retical and simulation model in this experimental setup. The RBDO problem has consid-
Figure 15.
15. AAphoto
photoof of
the practical experimental setup: (1) compliant mechanism; (2) PZT;(2)
(3) PZT;
laser
Figure
ered all variations in anthe practical
authentic experimental
manufacturing setup: (1) compliant
environment. mechanism;
sensor
(3) laserLK-G30; (4) translation
sensor LK-G30; stage (5)
(4) translation NI(5)
stage MyRio 1900;1900;
NI MyRio (6) vibration-isolated optical
(6) vibration-isolated platform;
optical (7)
platform;
controller LK-3001P; (8) driver PI E-470.20; (9) computer.
(7) controller LK-3001P; (8) driver PI E-470.20; (9) computer.

4.1. Natural
4.1. Natural Frequency
Frequency Evaluation
Evaluation
A continuously
A continuouslyvarying
varying frequency
frequencysine
sinewave
wave signal
signal excites
excites the
the PZT
PZT in
in the
the y-direction
y-direction
to determine
to determine the natural
natural frequency
frequency ofofthe
theCFDM.
[Link]
noncontactlaser
lasersensor
sensorrecords the
records y-
the
y-direction
direction displacement
displacement of of
thethe
mechanism.
mechanism. TheThe
displacement measurement
displacement measurement datadata
are ana-
are
analyzed
lyzed by by
fastfast Fourier
Fourier transform
transform (FFT),
(FFT), andand Figure
Figure 12b12b shows
shows thethe frequency
frequency response
response re-
sults. In Figure 12b, the frequency at the first peak, corresponding to the first resonant
frequency, is 1529.4 Hz. This measurement result is slightly larger than the simulation
result by FEM. This deviation is partly due to machining errors and imperfections in the
structure assembly.
 Appl. Sci. 2023, 13, 4968 21 of 27

results. In Figure 12b, the frequency at the first peak, corresponding to the first resonant
frequency, is 1529.4 Hz. This measurement result is slightly larger than the simulation
result by FEM. This deviation is partly due to machining errors and imperfections in the
structure assembly.

4.2. Control Experiment

Initially, the PZT actuator is supplied with a voltage, from 0 to 450 VDC, according to
the supplier’s instructions, to calibrate its unload displacement capability. The noncontact
displacement sensor LK-G30 measures the output displacement of the PZT. Figure 16 shows
the measurement results. This measurement result inherently indicates that the character
of the piezoelectric actuator is rate-independent hysteresis. This is the major limitation of
2023, 13, 4968 23 of 29
all piezoelectric actuators, which needs to be appropriately compensated for by the control
system to ensure good positioning accuracy.

(a)

(b)
Figure 16. UnloadFigure
PZT displacement measurement:
16. Unload PZT (a) measurement:
displacement setup and (b) measurement
(a) setup and data.
(b) measurement data.

To validate the To validate of

operability thethe
operability
mechanism, of the
an mechanism,
experimentalanopen-loop
experimental open-loop control
control
model, as shownmodel, as shown
in Figure in FigureTwo
14, is operated. 14, isnoncontact
operated. laser
Two noncontact
sensors, (1) laser sensors,
and (2), record(1) and (2), record
the output
the output displacement anddisplacement and parasitic
parasitic motion. [Link],
The LK-3001P The LK-3001P controller,
connected to a connected to a
computer via thecomputer
USB port,via the USB the
transmits port, transmits theresults.
measurement measurement [Link]-Naviga-
The software The software LK-Navigator
2 (Keyence, Osaka, Japan) stores and displays the measurement
tor 2 (Keyence, Osaka, Japan) stores and displays the measurement results on the results
com-on the computer.
puter. Figure 17 illustrates the open-loop displacement control results, which show that show that the
Figure 17 illustrates the open-loop displacement control results, which
the experimental displacement measurement does not exhibit a proportional relation as
does the theoretical one. Thus, a more precise control method is necessary.
Appl. Sci. 2023, 13, 4968 22 of 27

Appl.
Appl. Sci.
Sci. 2023,
2023, 13,
13, 4968
4968 24 of
of 29
24 does
experimental displacement measurement does not exhibit a proportional relation as 29
the theoretical one. Thus, a more precise control method is necessary.

Figure 17.
Figure 17. Open-loop
Open-loop displacement
displacement control
control results.
results.

shownininFigure
As shown Figure 15,15,
thethe
fullfull closed-loop
closed-loop controlcontrol
model model
uses anuses an additional
additional NI MyRio NI
1900
MyRio 1900
card to card to provide
provide feedback feedback
on theon the output
output displacement
displacement [Link].
This This
study study
usesuses
the
the GA-PID
GA-PID algorithm
algorithm in MATLAB
in MATLAB and and LabVIEW
LabVIEW to implement
to implement the closed-loop
the closed-loop control.
control. The
The single-objective
single-objective GA algorithm
GA algorithm makesmakes finding
finding the optimal
the optimal set ofset of PID
PID parameters
parameters [Link].
The
The values
values of those of parameters
those parameters
are K P = 𝐾 = 13.0166
are13.0166, , 𝐾 = 0.00753268
K I = 0.00753268, and K D = , and 𝐾 =
0.0019387.
0.0019387.
Figure Figurethe
18 shows 18relation
shows the relation
between thebetween
input and theoutput
input and output displacement
displacement of the closed-of
the closed-loop
loop control by control
the GA-PIDby the GA-PID algorithm,
algorithm, which is more which is more
linear thanlinear than17.
in Figure in Figure
Figure 17.
19
shows
Figure the set value,
19 shows the actual value,
set value, and value,
actual impulse response
and impulse when different
response wheninput waveforms
different input
(sine, triangle,
waveforms andtriangle,
(sine, square wave) are applied
and square wave)toaretheapplied
PZT actuator.
to the ThePZTresponses
actuator. have an
The re-
sponses have an error of less than 0.025 μm. This result shows that the control system has
error of less than 0.025 µm. This result shows that the control system has high accuracy
and
highresponsiveness. In comprehensive
accuracy and responsiveness. research, a more
In comprehensive detailed
research, analysis
a more of theanalysis
detailed current
PID control
of the current algorithm and algorithm
PID control procedure,and or application
procedure, of or other modern
application of control methods
other modern to
con-
evaluate
trol methods the performance
to evaluate the and effectiveness
performance of effectiveness
and the mechanism, should
of the be presented.
mechanism, shouldForbe
simplicity
[Link] For generality, thesegenerality,
simplicity and issues willthese
be discussed
issues will elsewhere,
be discussedas they are beyond
elsewhere, the
as they
scope of the current paper.
are beyond the scope of the current paper.

Figure 18.
Figure 18. Experimental
Experimental results
results of
of input/output
input/output displacement
displacement in
in closed-loop
closed-loop control.
control.
Appl.
[Link].
Appl. Sci. 2023,13,
Sci.2023,
2023, 13,4968
13, 4968
4968 23of
25
25 of 27
29
29

Figure19.
Figure [Link]
Experimental results
results of
ofclosed-loop
closed-loop control
control with
with impulse
impulse response:
response: (a)
(a) sine,
sine, (b)
(b) triangle,
triangle,
(c) square,
(c) square, and
square, and (d)
and (d) error
(d) error when
error when controlling.
when controlling.
controlling.
(c)

4.3. Mechanism
4.3. Mechanism Stiffness
Stiffness Evaluation
Stiffness Evaluation
Thetheoretical
The theoreticalmethod
methodhas hasderived
derivedthe
theinput
inputand
andoutput
outputstiffness
stiffnessof
stiffness ofthe
themechanism
mechanism
in
in Equations
Equations (21)
(21) and
and (22).
(22). This
This isis validated
validated using
using the
the FEMFEM method
method
in Equations (21) and (22). This is validated using the FEM method in Section 3.3. in in Section
Section [Link]. Figure
Figure
Figure20
20an
is
20 isexperimental
is an experimental
an experimental model
model
model for measuring
measuring
for measuring
for the output
the output
the output force
forceforce and input/output
input/output
and input/output
and displace-
displacement
displace-
of
menttheof
ment mechanism.
of the mechanism.
the A force
mechanism. gauge
A force
A force (NK-500)
gauge
gauge records
(NK-500)
(NK-500) the applied
records
records force,force,
the applied
the applied and and
force, the
andlaser dis-
the laser
the laser
placement
displacement sensor (LK-G30,
sensor (LK-G30, Keyence
Keyence Corp.,
Corp., Osaka,
Osaka,Japan)
Japan) measures
measures
displacement sensor (LK-G30, Keyence Corp., Osaka, Japan) measures the corresponding the
the corresponding
corresponding
displacement.
displacement. The
displacement. The experimental
The experimental stiffness
experimental stiffness is
stiffness is calculated
is calculated based
calculated based on
based on Hooke’s
on Hooke’s
Hooke’s law,law,
law,and
andis
and islisted
is listed
listed
in
in Table
Table
in Table 6. 6.
6.

Figure20.
Figure 20. Experimental
Experimentalmodel
modelto
tomeasure
measurethe
thestructure
structurestiffness.
structure stiffness.
stiffness.
Appl. Sci. 2023, 13, 4968 24 of 27

Table 6. Comparison of simulation and experiment results.

Performance Parameters Analytical Simulation Error (%) Experiment Error (%)

Amplification ratio 3 3.104 3.46% 2.93 2.33%
Natural frequency (Hz) 1519.2 1500 1.26% 1529.4 0.67%
Travel range (µm) 40 39.65 0.87% 39.16 2.1%
Input stiffnessKin (N/µm) 65.43 59.26 9.42% 62.35 4.70%
Output stiffness Kout (N/µm) 5.21 5.23 0.38% 5.42 4.03%

Table 6 presents the results of displacement amplification ratio, natural frequency,

and stiffness calculated using different methods, along with the experimental results. The
benchmarking among theory, simulation, and empirical models is determined using analyt-
ical results as a standard. The differences in results for the displacement amplification ratio
and natural frequency are 3.46% and 1.26%, respectively. The input and output stiffness
discrepancies between the experiment and theory are 4.7% and 4.03%, respectively. This
difference might affect the practical cutting forces with a given cutting depth. However, in
micromachining, cutting loads are relatively small, which moderately impacts the perfor-
mance of the instrument. The reason for this variance is imperfections in the fabrication
and assembly of the experiment and testing. Since this difference is modest, the proposed
theoretical model, simulation, and corresponding empirical models are adequate.
The results obtained for the CFDM design in this study are compared with some
previously published works, as shown in Table 7. With an amplification ratio of 3.0, the
current mechanism has a much more prominent first-natural frequency than previous
studies, and the concentrated stress is relatively small. These properties are beneficial
not only for high-precision feed drive mechanisms, but also for high-frequency vibration-
assisted machining applications using CMs.

Table 7. Comparison with the results of previous studies.

Natural Frequency
Studies Amplification Ratio Stress (MPa)
(Hz)
Wang [42] 398.8 1.21 53
Ding [43] 156.77 unknown 94.636
Kurniawan [44] 480 3 unknown
Li [45] and Liang [4] 200 2.8725 260.65
This study 1500 3.104 73.7

5. Conclusions
In the field of compliant mechanisms, this study has proposed a proper size optimiza-
tion method based on two steps: pseudo-rigid-body model optimization, and reliability-
based multi-objective optimization. The benchmarking for the optimization results between
the proposed RBDO and DO shows that all candidates in the Pareto front of the RBDO
satisfy the expected safety level, R0 . Meanwhile, not all optimum designs of the DO meet
the requirement. This optimization algorithm has been applied to design a new CFDM
with the reliability of R0 = 99.87% for two objective functions of natural frequency and
displacement amplification. It considers the uncertainty of different input/output parame-
ters of the problem, such as cutting forces, fabrication variations, material properties, and
input loads from the PZT actuator. This CFDM has been fabricated, tested, and controlled
using the GA-PID algorithm. The experimental results show that the discrepancies between
theory, simulation, and experiment are less than 5% for the first natural frequency and
displacement amplification ratio. Due to the symmetrical design of the mechanism, the
parasitic motion is decoupled from the instrument’s end-effector. In addition, the error of
displacement closed-loop control is 0.025 µm. This CFDM is also being used to investigate
the machining efficiency of microfabrication. The experimental results will be presented in
another published work.
Appl. Sci. 2023, 13, 4968 25 of 27

Author Contributions: Conceptualization, V.-K.N. and H.-T.P.; methodology, V.-K.N. and H.-H.P.;
software, H.-T.P.; validation, V.-K.N.; formal analysis, V.-K.N. and H.-T.P.; investigation, V.-K.N.;
resources, P.S.M. and Q.-K.D.; data curation, V.-K.N. and Q.-K.D.; writing—original draft preparation,
V.-K.N.; writing—review and editing, H.-T.P. and H.-H.P.; visualization, H.-T.P.; supervision, H.-T.P.;
project administration, H.-T.P.; funding acquisition, V.-K.N. and H.-T.P. All authors have read and
agreed to the published version of the manuscript.
Funding: This research was supported by the Ho Chi Minh City University of Technology and
Education, project grant number B2021-SPK-02, funded by the Ministry of Education and Training,
and hosted by the Ho Chi Minh City University of Technology and Education, Vietnam.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

Nomenclature
ADPL Ansys parametric design language
CFDM Compliant micropositioning feed drive mechanism
CM Compliant mechanism
DO Determined optimization
DOF Degree-of-freedom
DV Design variable
EDM Electrical discharge machining
FEM Finite element method
FORM First-order reliability method
MCS Monte Carlo simulation
NSGA-II Nondominated sorting genetic algorithm II
PRBM Pseudo-rigid-body model
PZT Piezoelectric actuator
RBDO Reliability-based design optimization

References
1. Howell, L.L. Compliant Mechanisms; John Wiley & Sons: New York, NY, USA, 2001.
2. Liu, Q.; Zhou, X.; Xu, P.; Zou, Q.; Lin, C. A flexure-based long-stroke fast tool servo for diamond turning. Int. J. Adv. Manu Tech.
2012, 59, 859–867. [CrossRef]
3. Nguyen, V.K.; Pham, H.T.; Pham, H.H.; Dang, Q.K. Optimization design of a compliant linear guide for high-precision feed drive
mechanisms. Mech. Mach. Theo. 2021, 165, 104442. [CrossRef]
4. Liang, H. A large-stroke flexure fast tool servo with new displacement amplifier. In Proceedings of the IEEE International
Conference on Advanced Intelligent Mechatronics (AIM), Munich, Germany, 3–7 July 2017. [CrossRef]
5. Watanabe, S.; Ando, T. High-speed XYZ-nanopositioner for scanning ion conductance microscopy. App. Phys. Lett. 2017,
111, 113106. [CrossRef]
6. Noveanu, S.; Lobontiu, N.; Lazaro, J.; Mandru, D. Substructure compliance matrix model of planar branched flexure-hinge
mechanisms: Design, testing and characterization of a gripper. Mech. Mach. Theory 2015, 91, 1–20. [CrossRef]
7. Chi, I.T.; Chanthasopeephan, T.; Wang, D.A. Design of a parallel gripper based on topology synthesis and evolutionary
optimization. J. Mech. Rob. 2021, 14, 021008. [CrossRef]
8. Pham, H.T.; Wang, D.A. A constant-force bistable mechanism for force regulation and overload protection. Mech. Mach. Theo.
2011, 46, 899–909. [CrossRef]
9. Han, Y.M.; Han, C.; Kim, W.H.; Seong, H.Y.; Choi, S.B. Control performances of a piezoactuator direct drive valve system at high
temperatures with thermal insulation. Smart Mat. Struc. 2016, 25, 097003. [CrossRef]
10. Ling, M.; Cao, J.; Zeng, M.; Lin, J.; Inman, D.J. Enhanced mathematical modeling of the displacement amplification ratio for
piezoelectric compliant mechanisms. Smart Mat. Struct. 2016, 25, 075022. [CrossRef]
11. Zhu, X.; Xu, X.; Wen, Z.; Ren, J.; Liu, P. A novel flexure-based vertical nanopositioning stage with large travel range. Rev. Sci. Inst.
2015, 86, 105112. [CrossRef]
12. Kim, C.J.; Kota, S.; Moon, Y.M. An instant center approach toward the conceptual design of compliant mechanisms. J. Mech. Des.
2005, 128, 542–550. [CrossRef]
Appl. Sci. 2023, 13, 4968 26 of 27

13. Kim, C.J.; Moon, Y.M.; Kota, S. A building block approach to the conceptual synthesis of compliant mechanisms utilizing
compliance and stiffness ellipsoids. J. Mech. Des. 2008, 130, 022308. [CrossRef]
14. Bernardoni, P.; Bidaud, P.; Bidard, C.; Gosselin, F. A new compliant mechanism design methodology based on flexible building
blocks. In Proceedings of the SPIE Smart Structures and Materials: Modeling, Signal Processing, and Control, San Diego, CA,
USA, 14–18 March 2004. [CrossRef]
15. Gallego, J.; Herder, A.J. Synthesis methods in compliant mechanisms: An overview. In Proceedings of the International
Design Engineering Technical Conferences and Computers and Information in Engineering Conference, San Diego, CA, USA,
30 August–2 September 2009. [CrossRef]
16. Bilancia, P.; Berselli, G. Design and testing of a monolithic compliant constant force mechanism. Smart Mat. Struct. 2020,
29, 044001. [CrossRef]
17. Nguyen, V.K.; Pham, H.T.; Pham, H.H. Optimal design of high precision compliant guide mechanism using gene algorithm and
Taguchi-based sensitivity analysis. In Proceedings of the International Conference on System Science and Engineering (ICSSE),
Ho Chi Minh City, Vietnam, 21–23 July 2017. [CrossRef]
18. Tian, Y.; Ma, Y.; Wang, F.; Lu, K.; Zhang, D. A novel XYZ micro/nano positioner with an amplifier based on L-shape levers and
half-bridge structure. Sens. Actuators A Phys. 2020, 302, 111777. [CrossRef]
19. Kim, H.; Kim, J.; Ahn, D.; Gweon, D. Development of a nanoprecision 3-DOF vertical positioning system with a flexure hinge.
IEEE Trans. Nanotech. 2013, 12, 234–245. [CrossRef]
20. Wang, F.; Liang, C.; Tian, Y.; Zhao, X.; Zhang, D. Design and control of a compliant microgripper with a large amplification ratio
for high-speed micro manipulation. IEEE/ASME Trans. Mechatron. 2016, 21, 1262–1271. [CrossRef]
21. Dang, M.P.; Le, H.G.; Van, M.N.; Chau, N.L.; Dao, T.-P. Modeling and optimization for a new compliant 2-dof stage for locating
biomaterial samples by an efficient approach of a kinetostatic analysis-based method and neural network algorithm. Comput.
Intell. Neurosci. 2022, 2022, 6709464. [CrossRef]
22. Tang, H.; Li, J.; Jia, Y.; Gao, J.; Li, Y. Development and testing of a large-stroke nanopositioning stage with linear active disturbance
rejection controller. IEEE Trans Autom. Sci. Eng. 2021, 19, 2461–2470. [CrossRef]
23. Song, L.K.; Fei, C.W.; Wen, J.; Bai, G.C. Multi-objective reliability-based design optimization approach of complex structure with
multi-failure modes. Aerospace Sci. Technol. 2017, 64, 52–62. [CrossRef]
24. Choi, S.K.; Canfield, R.A.; Grandhi, R.V. Reliability-Based Structural Optimization; Springer: Berlin/Heidelberg, Germany, 2007.
25. Nowad, A.S.; Collins, K.R. Reliability of Structure; McGraw-Hill: New York, NY, USA, 2000.
26. Li, F.; Wu, T.; Badiru, A.; Hu, M.; Soni, S. A single-loop deterministic method for reliability-based design optimization. Eng. Opt.
2013, 45, 435–458. [CrossRef]
27. Wu, J.; Yan, S.; Li, J.; Gu, Y. Mechanism reliability of bistable compliant mechanisms considering degradation and uncertainties:
Modeling and evaluation method. App. Math. Model. 2016, 40, 10377–10388. [CrossRef]
28. Wan, L.; Chen, H.; Ouyang, L.; Chen, Y. A new ensemble modeling approach for reliability-based design optimization of
flexure-based bridge-type amplification mechanisms. Int. J. Adv. Manuf. Technol. 2020, 106, 47–63. [CrossRef]
29. Zhan, Z.; Zhang, X.; Zhang, H.; Chen, G. Unified motion reliability analysis and comparison study of planar parallel manipulators
with interval joint clearance variables. Mech. Mach. Theory 2019, 138, 58–75. [CrossRef]
30. Guo, F.; Sun, Z.; Zhang, S.; Cao, R.; Li, H. Optimal design and reliability analysis of a compliant stroke amplification mechanism.
Mech. Mach. Theory 2022, 171, 104748. [CrossRef]
31. Zhan, Z.; Zhang, X.; Jian, Z.; Zhang, H. Error modelling and motion reliability analysis of a planar parallel manipulator with
multiple uncertainties. Mech. Mach. Theory 2018, 124, 55–72. [CrossRef]
32. Koseki, Y.; Tanikawa, T.; Koyachi, N.; Arai, T. Kinematic analysis of a translational 3-dof micro-parallel mechanism using the
matrix method. Adv. Robot. 2002, 16, 251–264. [CrossRef]
33. Lobontiu, N. Compliant Mechanisms: Design of Flexure Hinges; CRC Press: Boca Raton, FL, USA, 2020.
34. Young, W.C.; Budynas, R.G.; Sadegh, A.M. Roark’s Formulas for Stress and Strain; McGraw-Hill: New York, NY, USA, 2002.
35. Paros, J.M. How to design exural hinges. Mach. Des. 1965, 37, 151–156.
36. Pham, H.H.; Chen, I.M. Stiffness modeling of flexure parallel mechanism. Precis. Eng. 2005, 29, 467–478. [CrossRef]
37. Lobontiu, N. Compliance-based matrix method for modeling the quasi-static response of planar serial flexure-hinge mechanisms.
Precis. Eng. 2014, 38, 639–650. [CrossRef]
38. Wu, Y.; Zhou, Z. Design calculations for flexure hinges. Rev. Sci. Inst. 2002, 73, 3101–3106. [CrossRef]
39. Zhu, Z.; Zhou, X.; Liu, Z.; Wang, R.; Zhu, L. Development of a piezoelectrically actuated two-degree-of-freedom fast tool servo
with decoupled motions for micro-/nanomachining. Precis. Eng. 2014, 38, 809–820. [CrossRef]
40. Zhu, W.L.; Zhu, Z.; Shi, Y.; Wang, X.; Guan, K.; Ju, B.F. Design, modeling, analysis and testing of a novel piezo-actuated XY
compliant mechanism for large workspace nano-positioning. Smart Mat. Struct. 2016, 25, 115033. [CrossRef]
41. Tang, X.; Chen, I.M.; Li, Q. Design and nonlinear modeling of a large-displacement XYZ flexure parallel mechanism with
decoupled kinematic structure. Rev. Sci. Inst. 2006, 77, 115101. [CrossRef]
42. Wang, W.C.; Lee, J.W.; Chen, K.S.; Liu, Y.H. Design and vibration control of a notch-based compliant stage for display panel
inspection applications. J. Sound Vib. 2014, 333, 2701–2718. [CrossRef]
43. Ding, B.; Yang, Z.X.; Zhang, G.; Xiao, X. Optimum design and analysis of flexure-based mechanism for non-circular diamond
turning operation. Adv. Mech. Eng. 2017, 9, 168781401774335. [CrossRef]
Appl. Sci. 2023, 13, 4968 27 of 27

44. Kurniawan, R.T.; Ko, J. A study of surface texturing using piezoelectric tool holder actuator on conventional CNC turning. Int. J.
Precis. Eng. Manuf. 2013, 14, 199–206. [CrossRef]
45. Li, J. A Stable Autoregressive Moving Average Hysteresis Model in Flexure Fast Tool Servo Control. IEEE Trans. Autom. Sci. Eng.
2019, 16, 1484–1493. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.