Académique Documents
Professionnel Documents
Culture Documents
CHAPTER 1
INTRODUCTION
1.1
macro models can be used [6]. On the other hand, at the component or device levels
the physical behavior of three-dimensional continuums is described by partial
differential equations (PDE) solvable by Finite Element or Finite Difference
Element Methods (FEM or FDM)[3]. Micro-Electro-Mechanical Systems consists of
mechanical elements, sensors, actuators and electrical and electronics devices on a
common silicon substrate [7]. The sensors in MEMS gather information from the
environment through measuring mechanical, thermal, biological, chemical, optical,
and magnetic phenomena. The electronics then process the information derived from
the sensors and through some decision making capability, direct the actuators to
respond by moving, positioning, regulating, pumping and filtering, thereby
controlling the environment for some desired outcome or purpose [3].
The advantages of semiconductor IC manufacturing such as low cost,
mass production, reliability are also integral to MEMS devices. The size of MEMS
sub-components is in the range of 1 to 100 micrometers and the size of MEMS
device itself measures in the range of 20 micrometers to a millimeter [8]. These have
been used as sensors for pressure, temperature, mass flow, velocity, sound and
chemical composition, as actuators for linear and angular motions and as simple
components for complex systems such as robots, lab-on-a-chip, micro heat engines
and micro heat pumps[6]. The Lab on-a-chip in particular is promising to automate
biology and chemistry. To some extent the integrated circuit has allowed large-scale
automation of computation.
Accelerometers for automobile airbags, keyless entry systems, dense
arrays of micro mirrors for high definition optical displays, scanning electron
microscope tips to image single atoms, micro heat exchangers for cooling of
electronic circuits, reactors for separating biological cells, blood analyzers and
pressure sensors for catheter tips are but a few of the current usages. Micro ducts are
used in infrared detectors, diode lasers, miniature gas chromatographs and highfrequency fluidic control systems. Micro pumps are used for ink jet printing,
environmental testing and electronic cooling. Potential medical applications for
small pumps include controlled delivery and monitoring of minute amount of
medication, manufacturing of nanoliters of chemicals, and development of artificial
pancreas [5]. Commercial tools like MEMCAD (Microcosm Technologies) [7] and
MEMS Modeler (MEMSCAP) use parametric curve-fitting of simulation data to
obtain macro models [9]. The primary drawback of these methods is that they do not
generate scalable macro models.
However, the greatest potential for MEMS devices lies in new
applications within telecommunications (optical and wireless), biomedical and
process control areas. Military use of MEMS such as triggers for weapons, microgyros, micro-surety systems, and micro-navigation devices give another dimension
to the importance of reliability of these devices [8]. Any accidental triggering may
claim many lives and, if in a warehouse, may have a domino effect. Initial air bag
technology used conventional mechanical ball and tube type devices which were
relatively complex, weighed several pounds and cost several hundred dollars. They
were usually mounted in the front of the vehicle with separate electronics near the
airbag. MEMS have enabled the same function to be accomplished by integrating
an accelerometer and the electronics into a single silicon chip. Another example of
an extremely successful MEMS application is the miniature disposable pressure
sensor used to monitor blood pressure in hospitals. These sensors connect to a
patients intravenous (IV) line and monitor the blood pressure through the IV
solution. For a fraction of their cost ($10), the hospitals have replaced the early
external blood pressure sensors with MEMS. The early ones cost over $600 and had
to be sterilized and recalibrated for reuse [3].
MEMS have several distinct advantages as a manufacturing technology
[6]. In the first place, the interdisciplinary nature of MEMS technology and its
micromachining techniques, as well as its diversity of applications has resulted in an
unprecedented range of devices and synergies across previously unrelated fields (for
example biology and microelectronics). Secondly, MEMS with its batch fabrication
techniques enables components and devices to be manufactured with increased
performance and reliability, combined with the obvious advantages of reduced
physical size, volume, weight and cost. Thirdly, MEMS provides the basis for the
manufacture of products that cannot be made by other methods. These factors make
MEMS potentially a far more pervasive technology than integrated circuit
like, computer games, cell phones, pagers, PDAs, advanced robotics, laptop
computers, computer input devices, camcorders, digital cameras and SD card
accessories. Size and accuracy are the most essential features that decide the
performance of the sensor in each of the above-mentioned applications [11].
In recent years, CMOS micromachining has evolved as a chief
fabrication technology for VLSI MEMS. CMOS micromachining is utilized for
fabricating the design and characterization of a lateral accelerometer. More attention
is being paid towards integrated Microsystems, such as inertial measurement unit.
Integrated Microsystems yield improved performance with an array of similar
topology devices, like accelerometers and gyroscopes, which possess dissimilar
performance specifications [12].
1.2
INTRODUCTION TO OPTIMIZATION
But ask the animals, and they will teach you, or the birds of the air, and
they will tell you; or speak to the earth, and it will teach you, or let the fish of the sea
inform you.
-Job 12:7 (adapted from [179])
Optimization is the procedure or procedures used to make a system or
design, as effective or functional as possible, especially the mathematical techniques
involved; making the best of anything. It is a mathematical technique used for
finding a maximum or minimum value of a function for several variables subject to
a set of constraints, as linear programming or system analysis.
1.3
TYPES OF OPTIMIZATION
The types of optimization techniques are Unconstrained optimization,
1.3.1
Unconstrained Optimization
Unconstrained optimization problem is one where you only have to be
concerned about the objective function you are trying to optimize. None of the
variables in the objective function are constrained.
1.3.2
Constrained Optimization
Constrained optimization is the process of optimizing an objective
Multi-Objective Optimization
Multiobjective optimization also known as multi-objective programming,
Multimodal Optimization
Multimodal optimization problem is a problem that has more than one
local minimum or multimodal optimization deals that with optimization takes, that
involve finding all or most of the multiple solutions (as opposed to a single best
solution) .
1.3.5
Combinatorial Optimization
Hill Climbing
Hill climbing is a graph search algorithm where the current path is
extended with a successor node which is closer to the solution than the end of the
current path. In simple hill climbing, the first closer node is chosen whereas in
steepest ascent hill climbing all successors are compared and the closest to the
solution is chosen.
1.3.7
Intelligence
Intelligence has been defined in many different ways such as in terms of
design of MEMS. The Table 1.1 gives the details of various classic evolutionary
algorithms developed for optimization.
Author
Reference
Dorigo
[13]
[16]
Li et al.
[23]
Passino
[25,26]
Yang
[28]
[42]
John Holland
[50]
Simon
[58]
[66]
He et al.
[69]
[71]
Rashedi et al.
[79]
Geem et al.
[97]
Kirkpatrick et al.
[111]
Other Algorithms
Cultural Algorithm (2004)
Reynolds
[114]
Larranaga
[117]
Tizhoosh
[124]
[130]
Rao
[134]
1.4.1
ACO initially proposed by Marco Dorigo in 1992 in his Ph.D thesis [13].
Ant colony optimization is a way to solve optimization problems based on the way
ants indirectly communicate directions to each other. Ant colony optimization is a
probabilistic technique for solving computational problems which can be reduced to
find good paths through graphs. Each ant tries to find a route between its nest and a
good food source [14].
The behavior of each ant in nature:
(i)
(ii)
(iii)
(iv)
However pheromones evaporate over time, such that unless they are
reinforced by ants, the pheromones will disappear. Applications of ACO includes
scheduling problems(project scheduling, job shop scheduling, open shop, agent
based dynamic scheduling), routing problems (TSP, vehicle routing, connection
oriented and connection less network routing), assignment problems(quadratic
assignment problems, course timetabling, graph coloring),sequential ordering
problem, shortest common super sequence problem, constraint satisfaction,
classification rules, bayesian networks, protein folding, protein-ligand docking, set
problem and device sizing problem in nanoelectronics physical design, machine
learning, dynamic problem of data network routing, a shortest path problem where
properties of the system such as node availability vary over time, continuous
optimization and parallel processing implementations, digital image processing and
classification problem in data mining[15].
1.4.2
10
11
12
The firefly algorithm was introduced in the year 2007 by Xin-She Yang
at Cambridge University [28]. It is based on the attraction of fireflies to one another.
Attraction is based on the perceived brightness of a firefly, which exponentially
decreases with distance. A firefly is attracted only to those fireflies that are brighter
than it. Firefly can be described by using following idealized rules
(i)
All fireflies are unisex so that one firefly will be attracted to other
fireflies regardless of their sex.
(ii)
13
1.4.6
14
John Holland in early 1970s [50]. Genetic Algorithms are inspired by Darwins
theory about evolution. Solution to a problem by Genetic Algorithms evolved.
Genetic Algorithms and evolutionary strategies mimic the principle of natural
genetics and natural selection to construct and search optimization procedures. GA
belong to the larger class of EA which generate solutions to optimization problems
using technique inspired by natural evolution, such as inheritance, mutation,
selection and crossover.
Algorithm is started with a set of solutions (represented by
chromosomes) called population. Solutions from one population are taken and used
to form new population. This is motivated by a hope, that the new population will be
better than the old one. Solutions which are selected to form new solutions
(offspring) are selected according to their fitness. The more suitable they are, the
more chances they have to reproduce. This is repeated until some condition is
satisfied [51, 52].
GA can even be faster in finding global maxima than conventional
methods, in particular when derivatives provide misleading information. The
15
Domain
Control
Design
Scheduling
4.
Robotics
Trajectory planning
Machine
Learning
1.4.8
others
Application types
Reference
[53-57]
science and study of species movement from one habitat to another. BBO is an
16
Immigration
E
Emigration
Rate
Sc Smax
Number of species
Figure 1.1 BBO Migration Curves (Adapted from [179])
In the Figure 1.1 the immigration rate and the emigration rate are the
functions of the number of species in a habitat. Here each habitat is a candidate
solution to an optimization problem, each species is an independent variable of that
candidate solution.
17
In BBO, each candidate solution shares its features with other candidate
solutions, and this sharing process is analogous to migration in biography. If
migration occurs for many iterations, the habitat becomes more suitable for their
species, which corresponds to candidate solutions providing increasingly better
solution to an optimization problem.
Application of BBO includes Economic Load Dispatch (ELD) problem
[59] with generator constraints in power plants. BBO has also solved real-world
application problems such as Block based motion estimation in video coding [60],
Implementing color image segmentation [61], color image quantization [62], Face
recognition [63], Feature selection [64], ECG signal classification, power system
optimization, ground water detection, and satellite image classification [65], general
benchmark functions, constrained optimization, the sensor selection problem for
aircraft engine health estimation, web based BBO graphical user interface, global
numerical optimization, and optimal meter placement for security constrained state
estimation [24].
1.4.9
18
19
based on the law of gravity and the notion of mass interactions. The GSA algorithm
20
uses the theory of Newtonian physics and its searcher agents are the collection of
masses. In GSA, we have an isolated system of masses. Using the gravitational force
every mass in the system can see the situation of other masses. The gravitational
force is therefore a way of transferring information between different masses. In
GSA, agents are considered as objects and their performance is measured by their
masses.
All these objects attract each other by a gravity force, and this force
causes a movement of all objects globally towards the objects with heavier masses.
The heavy masses correspond to a good solution of the problem. The position of the
agent corresponds to a solution of the problem, and its mass is determined using a
Fitness function (F) [79].
By lapse of time, masses are attracted by the heaviest mass. We hope that
this mass would present an optimum solution in the search space. The GSA could be
considered as an isolated system of masses. It is like a small artificial world of
masses obeying the Newtonian laws of gravitation and motion.
Application of GSA includes in the following fields: NN training [80],
robotics [81], optical [82], bioinformatics [83], software engineering [84],
networking [85], image processing [86], classification [87], clustering [88],
scheduling [89], business [90], computer engineering [91], civil engineering [92],
control engineering [93], mechanical engineering [94], power engineering [95],
telecommunication engineering [96].
1.4.13
21
insight in to optimization problems and can suggest efficient algorithms for solving
them.
-Valdo Cerny 1985 (adapted from [179] )
Simulated annealing was independently described by Scoot Kirkpatrick,
C. Daniel Gelatt and Mario P. Vecchi in 1983 [111]. Simulated annealing is an
optimization algorithm that is based on the cooling and crystallizing behavior of
chemical substances. Simulated annealing is a single individual stochastic algorithm.
Simulated annealing mimics the cooling phenomenon of molten metals to constitute
a search procedure. Slowly cool down a heated solid, so that all particles arrange in
the ground energy state. At each temperature wait until the solid reaches its thermal
equilibrium. Probability of being in a state with energy E can be represented by the
Equation (1.1),
Pr E = E =
(-E / KB).T
1
e
Z(T)
(1.1)
22
where,
E
Energy
Temperature
Boltzmann constant
KB
Z (T) -
Normalization factor
23
Vote:
Promote:
Acceptance
Influence
function
function
Inherit
Population Space
Reproduce, Modify,
Variation
Performance
Function
24
Tabu Search
Tabu search was introduced in 1986 by Glover and McMillan [130]. It is
25
method is still actively researched, and is continuing to evolve and improve. Tabu or
Taboo means forbidden, banned, or not allowed. Forbidden items, speech, or
practices can be based on culture, religion, morality or politics. Tabu search is a
higher level heuristic procedure for solving optimization problems, design to guide
other methods to escape the trap of local optimality [131,132]. Tabu search has
obtained optimal and near optimal to a wide variety of classical and practical
problems in application ranging from scheduling to telecommunications and from
character recognition to neural networks.
Applications of Tabu search includes employee scheduling, maximum
satisfiability problems, character recognition, space planning and architectural
development, telecommunications path assignment, probabilistic logic problems,
NN pattern recognition, quadratic assignment problems, network topology design,
computer channel balancing, TSP, graph coloring, graph partitioning, nonlinear
covering, maximum stable set problems, flow shop sequencing, design, location and
allocation, logic and artificial intelligence, technology, general combinational
optimization, graph optimization, routing, production, inventory and investment
[133].
1.4.19
26
1.5
RESEARCH MOTIVATION
Motivation of this research work is to optimize the parameters L1, L2, L3,
ym and Fitness function (F) or Die Area (DA) values of a MEMS accelerometer
using Artificial Bee Colony (ABC) algorithm with Particle Swarm Optimization
algorithm methods. The main problem in MEMS is to get an optimal design. The
significance of MEMS optimization relating to concert, power utilization, and
consistency increases. There are various optimization algorithms that can produce an
optimized design of MEMS.
Some of the optimization algorithms are methods without derivatives
(e.g. Nelder-Mead-Simplex), methods using derivatives (e.g. Conjugate Gradient or
Quasi-Newton). These methods have various disadvantages like, with an rising
amount of parameters these surfaces turn out to be more and more problematical and
it is almost not possible to compute them.
An additional universal difficulty is the typically complex relationship
among structure parameters and the structure performance. Probably the most
important drawback is finding a global optimum. FEM simulation is also another
important method to achieve the optimized design of MEMS. Here a simplified
spring-mass model is used to predict the device sensitivity. This method also will not
give a efficient optimized design of MEMS. The spring constant of the beams
should be further reduced by using some more complaint flexure structures (e.g.
four-fold beam) to get a better optimized method. All the drawbacks mentioned
above can be overcome by means of using a Genetic Algorithm. By using Genetic
Algorithm, MEMS design can be optimized in a better way. But GA having some
draw backs such as representation, population size and mutation rate, selection and
deletion policies, crossover and mutation operators, termination criterion. The
disadvantages of GA can be overcome by means of using ABC algorithm.
In general, the cost and the Die Area (DA) of the accelerometer are
directly proportional. Thus, the cost associated with the design of accelerometer
increases slowly with the increase in Die Area (DA) of the accelerometer. This
27
behavior has lead to the need for minimizing the Die Area (DA) along the design
parameter force (N) and thereby, optimization of these parameters comes into effect.
The optimization algorithm is centered on objective function or Fitness function (F).
The proposed method uses a combination of Artificial Bee Colony (ABC)
optimization algorithm and Particle Swarm Optimization (PSO) algorithm to
optimize the design parameters of MEMS accelerometer.
1.6
RESEARCH OBJECTIVES
MEMS based accelerometers used for airbag deployment in automobile
28
LITERATURE REVIEW
The various works related to the MEMS design optimization has been
presented here.
John K. Sakellaris [144] has presented the design of a vibration control
mechanism for a beam with bonded piezo electric sensors, actuators and an
application of the arising smart structure for vibrations suppression too. The
mechanical modeling of the structure and the subsequent finite element
approximation were based on Hamiltons principle and classical engineering theory
of bending of beams in connection with simplified modeling of piezoelectric sensors
and actuators. Two control schemes LQR and H2 were considered.
Aniket Singh et al. [145] have presented a study of MEMS RF Power
Sensors. An optimized sensor with low reflection loss parameters was identified.
Two designs, one with a cantilever bridge and another with a fixed bridge were
compared in terms of reflection and transmission losses. The designs were simulated
with different dielectric layers and varied thickness to get a series of results. A fairly
optimized design was realized with minimum reflection losses.
Xiaolin Chen et al. [146] have presented a study on multi-level
simulation that proved to be an efficient way to accelerate the design process and
improve the device performance. It can be used effectively to optimize the
29
gyroscope system. The device design was automatically generated based on mask
layout and fabrication process restrictions. Design verification was performed at the
device-level for detailed analysis and at the system-level for behaviour
characterization.
Chaitanya Chandrana et al. [147] have presented the structure for an
integrated transducer that used a non-conductive epoxy for mechanical backing of
the transducer and a thin film electrode for backside contact as part of the
integrated process for the transducer. The desired outcome was a single integrated
MEMS PVDF transducer chip, combining a high input impedance preamplifier and
focused transducer. It showed an approach for building integrated PVDF transducers
with minimal parasitic that could be widely used in clinical IVUS applications.
Adam Dugosz [148] has presented the MOOPTIM algorithm that had
been used for multiobjective shape optimization of MEMS structures. The
effectiveness of MOOPTIM had been compared to NSGAII on several benchmark
test problems. The obtained results showed the effectiveness of MOOPTIM for both
un-constrained and constrained optimization tasks. To reduce the time of the
optimization, parallel computation or approximate surrogate evaluations were used.
Rohit Pathak and Satyadhar Joshi [149] have presented a novel way to
approach reliability calculations and shown how properties at different levels and
types needed to be linked up in a multi scale analysis, where HPC can benefit
reliability calculations for MEMS devices. They have calculated various parameters
of different scale for a MEMS device and proceeded with the analysis of reliability
using MATLAB distributive computing Toolbox.
Sujata N. Naduvinamani et al. [150] have demonstrated the design of
cantilever based switch using CoSolve-EM and it was observed that pull in voltage
for RF MEMS switch varies for different dimensions. They have recently developed
CoSolve-EM, a coupled solver for 3D quasi-static electro-mechanics. With the help
of CoventorWare, the switch was designed. Initially in the process editor,
required gap (i.e between the beam and the substrate) was set to some desired
30
value. Then the 2-D layout was drawn. Then with the layout editor, the 3-D layout
was drawn. Then the meshing of the structure was done. After meshing the MEMMech set up was done for the CoSolve-EM and existing loads were removed. It was
solved for the CoSolve-EM and the mechanical deflections that takes place
solely due to the electrostatic force were observed. Then different voltages were
applied and the corresponding deflections towards the substrate were noted down.
Prince Nagpal et al. [151] have described the capacitive pressure
sensor design for biomedical applications like blood pressure measurement. The
described pressure sensors provided high sensitivity even at low pressure. This
makes it suitable for biomedical applications. Effects of varying different parameters
on the pressure sensor performance have been studied. From the results, the pressure
sensors with compatible parameters can be selected for specific requirements. These
compact pressure sensors are made up of biocompatible materials and can be
implanted easily inside body to be used for RF telemetry purpose.
Shveta Jain et al. [152] have presented the Performance Study of RF
MEMS Ohmic Series Switch. The effect of different geometrical parameters was
studied and simulated using CoventorWare. It showed that varying the anchors
length improves the contact force thereby reducing the insertion loss. Thus the tradeoff between the parameters and switch performance can be enhanced by
maintaining the parameters.
Zhang et al. [153] implemented a hierarchical MEMS synthesis and
optimization architecture, integrating an object-oriented data structure with SUGAR
and two types of optimization: Genetic Algorithms (GA) and local gradient-based
refinement. They noted that the MOGA approach needed a means for automating the
starting populations for MOGA that would enable a larger sampling of the solution
space of MEMS design.
Jain et al. [154] developed a MEMS transducer for an ultrasonic flaw
detection system. This experiment appears to be the first attempt to detect ultrasonic
signal by MEMS transducers in direct contact with solids.
31
32
ORGANIZATION OF THESIS
Chapter 1 discusses the overview of MEMS technology and its
33
SUMMARY
The significance of MEMS technology and its applications are discussed.
Among various optimization evolutionary algorithm techniques GA, ABC, PSO are
widely used, advantages and disadvantages of these optimization techniques are
discussed. The motivation for optimizing the parameter values of MEMS based
accelerometer and objectives of the research work are discussed. Literature reviews
on various works related to the MEMS design optimization are carried out.
Algorithms for optimization are identified.
CHAPTER 2
MEMS ACCELEROMETER DESIGN OPTIMIZATION
USING GENETIC ALGORITHM
2.1
INTRODUCTION
34
structure) may include the parameters like Beam length, Beam width, Beam depth,
Beam mass, proof mass etc. The parameters and its specifications are represented
below as follows,
Beam Length
Beam Width
Beam Depth
Beam Mass
a {xa, ya}
Proof Mass
m {xm, ym}
35
Lb O
(2.3)
ya P
(2.4)
x a Lb 2W1
(2.5)
x m Lb 2 L2 2W1 2W3
(2.6)
O
where
=150m, P=100m
L1 Q
(2.7)
L2 R
(2.8)
36
L3 S
(2.9)
ym T
(2.10)
To apply Genetic Algorithm, L1, L2, L3, ym assigned to Q, R, S, T using
Equation (2.7) to Equation (2.10), and we take random values for the design
variables within the following ranges (20m Q 500m, 20m R 100m,
100m S 500m, 100m T 500m) . These ranges have been chosen
based on the minimum size constraints and maximum area constraints, in addition to
general observation and intuition about the final designs optimal geometry [11]. The
parameters are optimized using the Genetic Algorithm. The process of Genetic
Algorithm is explained below.
2.3
GENETIC ALGORITHM
Genetic Algorithms are good at taking large, potentially huge search
spaces and navigating them, looking for optimal combinations of things, solutions
you might not otherwise find in a lifetime.
2.3.1
GA Definitions
A Genetic Algorithm is an iterative procedure maintaining a
37
some of the innovative flair of human search. In every generation, a new set of
artificial creatures (strings) is created using bits and pieces of the fittest of the old,
an occasional new part is tried for good measure. While randomized, Genetic
Algorithms are no simple random walk. They efficiently exploit historical
information to speculate on new search points with expected improved
performance [51].
2.3.2
Biological Background of GA
The science that deals with mechanisms responsible for similarities and
38
The cell
Every animal (or) human cell is a complex of the many small factories
that work together. The center of all this is the cell nucleus. The genetic information
is contained in the cell nucleus. The Figure 2.2 and Figure 2.3 show anatomy of the
animal cell and anatomy of animal nucleus [166].
39
Chromosomes
A set of genes. Chromosome contains the solution in the form of genes.
40
2.3.3.3
Reproduction
Reproduction of genetics is divided into two types. They are Mitosis and
Meiosis. Mitosis is copying the same genetic information to new offspring. There is
no exchange of information. It is the normal way of growing of multicell structures,
like organs. The Mitosis and Meiosis methods of reproduction are shown in
Figure 2.5.
Figure 2.5 Mitosis and Meiosis form of Reproduction (Adapted from [166])
41
42
2.3.3.4
Natural selection
The origin of species is based on preservation of favorable variations
and the rejection of unfavorable variations. There are more individual born than
that can survive, so there is a continuous struggle for life. Individuals with an
advantage have a greater chance to survive: survival of the fittest [166, 167].
2.3.4
Figure 2.8.
Population (Chromosomes)
Decoded String
Offspring
Parents
New generation
Genetic operations
CC
Evaluation
Calculation
Mate
Reproduction
Selection
43
Start
End
NO
Reproduce and
kill organisms
Mute organisms
44
2.3.4.1
Simple GA
The following basic program explains the simple Genetic Algorithm. GA
Various steps of GA
Step 1:
Step 2:
Step 3:
45
Step 4:
Step 5:
Step 6:
Step 7:
Step 8:
Repeat Step 5 until the size of the new chromosome population becomes
equal to the size of the initial population, N.
Step 9:
Step 10:
which define a proposed solution to the problem that the Genetic Algorithm is trying
to solve. The chromosome is often represented as a simple string; although a wide
variety of other data structures are also used.
In a Genetic Algorithm, a population of strings (called chromosomes),
which encodes candidates solutions (called individuals) to an optimization problem,
evolves toward better solutions. The design parameters that should be optimized are
specified by the following Equation (2.11),
K= {L1, L2, L3, ym}
(2.11)
46
Fitness function (F) should be evaluated by using Die Area (DA) Equation (2.16).
Based on the fitness, the four design parameter values are evaluated. The four design
parameters L1, L2, L3, ym of MEMS accelerometer should satisfy the following
conditions mentioned in Equation (2.12) to Equation (2.14),
L1 Q
(2.12)
L2 R
(2.13)
L3 W2 y a L1
(2.14)
If it does not satisfy the above conditions, the evaluated values will be
thrown out by giving a Fitness function (F) a value of infinity. If it satisfies, then the
values of Fitness function (F) are sorted to get 10 smallest values (parents).
A Fitness function (F) is a particular type of objective function that is
used to summarize, as a single figure of merit, how close a given design solution is
to achieving the set aims. The objective function or Fitness function (F) for the
MEMS design is represented by the Equation (2.15),
F D
(2.15)
D = (x m + 2L3 + 2W2 )y m
(2.16)
47
2.3.5
Crossover
A crossover is a genetic operator used to vary the programming of a
(I) i
(I)
+ (1- )i+1
(2.17)
i+1 =
(II) i
(II)
+ (1- )i+1
(2.18)
i
Where
and
(I)
are the children design values, i are the parent design values
(II)
are random values between zero and one. Thus the children design
variables can be produced from the parent design variables. Thus the obtained values
can be used for the further combination of random strings to create a new generation
or population [168].
2.3.6
Mutation
Mutation is a genetic operator used to maintain genetic diversity from
48
Selection
Selection is the stage of a Genetic Algorithm in which individual
for L1, L2, L3, ym and Fitness function (F) or Die Area (DA). 50 iterations were used
to obtain the optimal design. The iterative genetic process has minimized the Die
Area (DA), represented by the objective or Fitness function (F), by satisfying the
design criteria. The best performing design was saved for each successive starting
population to converge on the optimum values. The results had been displayed for
every 10 iterations the optimum values obtained by the Genetic Algorithm have been
shown in the Table 2.1 to Table 2.5. After 1,000 starting populations of 50
49
generations (iterations) have been computed, the best performing optimized design
parameters are given as the output.
After 10 iterations using the GA method the top 5 ranks are displayed in
the Table 2.1. Among these five rank values using rank selection method Rank 1
which produce the optimized parameter values L1= 2.826 x 10-04, L2= 8.454 x 10-05,
L3= 3.984 x 10-04, ym= 1.006 x 10-04, F= 113309.10072 m2. The minimized
objective function or Fitness function (F) at 10th iteration is F= 113309.10072
m2. Figure 2.10 represents the convergence of the objective function or Fitness
function (F) at 10th iteration.
Table 2.1 Five Optimal Values Obtained in the 10th Iteration by GA
Iteration
L1
L2
L3
ym
Rank 1
2.826e-04
8.454e-05
3.984e-04
1.006e-04
113309.10072
Rank 2
3.204e-04
4.655e-05
4.341e-04
1.011e-04
113378.55933
Rank 3
2.915e-04
8.160e-05
4.000e-04
1.010e-04
113508.45544
Rank 4
2.745e-04
8.714e-05
3.839e-04
1.033e-04
113899.40005
Rank 5
2.894e-04
6.075e-05
4.290e-04
1.001e-04
114141.35451
50
L1
L2
L3
ym
Rank 1
2.837e-04
8.604e-05
3.924e-04
1.010e-04
112903.54656
Rank 2
2.808e-04
9.071e-05
3.928e-04
1.004e-04
113180.96991
Rank 3
2.902e-04
9.063e-05
3.984e-04
1.000e-04
113920.92759
Rank 4
2.953e-04
6.791e-05
4.146e-04
1.014e-04
114188.22835
Rank 5
2.630e-04
9.141e-05
3.824e-04
1.031e-04
114232.92758
51
L1
L2
L3
ym
52
n
Rank 1
2.963e-04
6.568e-05
4.139e-04
1.001e-04
112072.49052
Rank 2
2.717e-04
9.759e-05
3.792e-04
1.014e-04
112972.01599
Rank 3
2.981e-04
6.259e-05
4.218e-04
1.002e-04
113230.17173
Rank 4
3.158e-04
5.038e-05
4.313e-04
1.012e-04
113824.23060
Rank 5
2.809e-04
8.586e-05
4.019e-04
1.003e-04
113917.35371
53
Iteratio
n
L1
L2
L3
ym
Rank 1
3.249e-04
4.827e-05
4.318e-04
1.006e-04
112754.92528
Rank 2
2.898e-04
8.056e-05
3.944e-04
1.016e-04
112892.53307
Rank 3
2.794e-04
9.760e-05
3.824e-04
1.008e-04
12922.54112
Rank 4
3.159e-04
5.661e-05
4.259e-04
1.005e-04
113114.38445
Rank 5
2.995e-04
6.010e-05
4.144e-04
1.019e-04
113129.17799
54
m2. Figure 2.14 represents the convergence of the objective function or Fitness
function (F) at 50th iteration.
Table 2.5 Five Optimal Values Obtained in the 50th Iteration by GA
Iteratio
n
L1
L2
L3
ym
Rank 1
3.085e-04
6.019e-05
4.107e-04
1.019e-04
112297.95163
Rank 2
3.108e-04
5.896e-05
4.198e-04
1.009e-04
112864.70542
Rank 3
3.061e-04
5.763e-05
4.127e-04
1.028e-04
113203.34803
Rank 4
2.832e-04
6.901e-05
4.114e-04
1.012e-04
113470.35106
Rank 5
2.972e-04
6.460e-05
4.223e-04
1.001e-04
113525.79582
L1=
55
Property Value
L1
L2
L3
ym
112297.95163 m2
SUMMARY
In this chapter, it has been discussed about the optimal design of MEMS
accelerometer. The parameters have been optimized to get the minimized Die Area
(DA) or Fitness function (F). Genetic Algorithm developed in MATLAB 7.12 had
been used for the optimization and it has produced a minimized Die Area (DA)
based on the Fitness function (F). Finally, optimal parameter values are obtained by
means of this algorithm. Thus, the experimental results have shown the optimal
design with minimized Die Area (DA) and optimized parameters.
56
CHAPTER 3
DESIGN PARAMETER OPTIMIZATION BASED ON
ARTIFICIAL BEE COLONY (ABC) ALGORITHM
FOR MEMS ACCELEROMETERS
3.1
INTRODUCTION
The parameters Z1, Z2, Z3, yk and Die Area (DA) or Fitness function (F) of
57
is carried out using the ABC algorithm and the final optimized parameters are
obtained as the final solution.
Optimization of design
Accelerometer
Figure 3.1 Block Diagram for Artificial Bee Colony (ABC) Optimization of
MEMS Accelerometer
3.3
with the aid of the various design parameters like Beam length, Beam width, and
Beam depth, Beam mass, proof mass etc. The parameters and its specifications are
represented below as follows,
Beam Length
Beam Width
Beam Depth
Beam Mass
m {xm, ym}
Proof Mass
k {xk, yk}
58
Among these parameters, Z1, Z2, Z3, yk should be optimized to produce the
optimal design of MEMS. The remaining parameters are assigned to the constant
values as follows using Equation (3.1) and Equation (3.2),
H1 H 2 H 3 H j A
(3.1)
D1 D2 D3 D j B
(3.2)
where A=B=1.8m.The Figure 3.2 MEMS accelerometer diagram specifies all the
parameters that can be used to get an optimal design of MEMS. It makes use of a
folded beam structure. The structure is specified as follows using Equation (3.3) to
Equation (3.6),
59
(3.3)
ym Q
(3.4)
x m Z j 2H 1
(3.5)
xk H j 2Z 2 2H 1 2H 3
(3.6)
where P=150m, Q=100m.
whereas the other parameters like Z1, Z2, Z3, yk assigned to R, S, T, U using
Equation (3.7) to Equation (3.10) .
Z1 R
(3.7)
Z2 S
(3.8)
Z3 T
(3.9)
yk U
(3.10)
We take random values for the design variables it should be in the
following ranges, where 20m R 500m, 20m S 100m, 100m T
500m, 100m U 500m [11]. The parameters are optimized using the ABC
algorithm. The process of ABC is explained below.
60
3.4
Colony (ABC) algorithm and different phases of ABC algorithm. A simple Artificial
Bee Colony algorithm was given as an example.
3.4.1
using the Figure 3.3 behavior of honey bee foraging for nectar. In the Figure 3.3 we
assume that there are two discovered food sources: A and B. At the very beginning, a
potential forager will start as unemployed forager and that forager bee will have no
knowledge about the food sources around the nest. There are two possible options
for such a bee:
61
Figure 3.3 Behavior of Honeybee Foraging for Nectar (Adapted from [169])
A, B
Onlooker
Scout
UF
Uncommitted Follower
EF1
Sharing Information
EF2
(i) It can be a scout and start searching around the nest spontaneously
for food due to some internal motivation or possible external clue
(S on Figure 3.3).
(ii)
After finding the food source, the bee utilizes its own capability to
memorize the location and then immediately starts exploiting it. And hence, the bee
becomes an employed forager. The foraging bee takes a load of nectar from the
source and return to the hive, thereby unloading the nectar to a food store. After
unloading the food, the bee has the following options:
(i)
(ii)
It might dance and then recruit nest mates before returning to the
same food source (EF1).
62
difference between the eventual total number of bees and the number of bees
presently foraging [169].
Artificial Bee Colony (ABC) is a novel optimization algorithm inspired
by the natural behavior of honey bees in their search process for the best food
sources [170].
In the ABC algorithm, the colony of artificial bees contains three groups:
employed bees, onlookers and scouts [171,172]. At the initialization step, a set of
food source positions is randomly produced and also the values of control
parameters of the algorithm are assigned [173]. The nectar amount retrievable from
food source corresponds to the quality of the solution represented by that food
source. So the nectar amounts of the food sources existing at the initial positions are
determined [174].
A bee waiting for the dance area to obtain the information about food
sources is called an onlooker, a bee going to the food source is named as an
employed bee, and a bee carrying out random search is called a scout [175]. The
goal of bees in the ABC model is to find the best solution. The position of a food
source represents a possible solution to the optimization problem and the nectar
amount of a food source corresponds to the quality (fitness) of the associated
solution [176].
After sharing their information with onlookers, every employed bee goes
to the food source area visited by itself at the previous cycle since that food source
exists in her memory, and then chooses a new food source by means of visual
information in the neighborhood of the one in its memory and evaluates its nectar
amount [169]. At the second stage after sharing the information, every employed
bee goes to the food source area visited by itself at the previous cycle since that food
source exists in her memory, and then chooses a new food source by means of visual
information in the neighborhood of the present one [177]. At the third stage, an
63
onlooker prefers a food source area depending on the nectar information distributed
by the employed bees in the dance area [179]. When the source is abandoned, the
employed bee becomes a scout and starts searching a new source in the vicinity of
the hive.
In ABC algorithm the possible solutions for the optimization problem can
be denoted by the position of a food source and the fitness of a related solution can
be obtained by the nectar quantity of food source. The quantity of solutions in the
population is equal to the quantity of the employed bees or the onlooker bees. At the
first cycle, ABC produces a randomly distributed initial population of solutions.
After initialization, the population of the solutions is focused to the
replicated cycles of the search process of the employed bees, the onlooker bees and
the scout bees. An employed bee generates alteration on the solution in their
memory depending on the area knowledge and verifies the nectar quantity of the
new solution. If the nectar quantity of the new one is better than that of the earlier
one, the bee memorizes the new position and forgets the old one or else it memorize
the earlier position in their memory. The employed bees, after their completion of
search process share the nectar knowledge of the food sources and their position
knowledge with the onlooker bees. Then onlooker bees estimate the knowledge from
the employed bees and select a food source with a probability similar to its quantity.
If it is satisfied then we can memorize the solution and stop the process, if not the
process has to be repeated until the new solution satisfies the eligibility criteria. The
working process of the ABC algorithm is based on the fitness value of the solution.
The Figure 3.4 shows the overall process that is being carried out in the
ABC algorithm. The first stage is the initialization of the population which is
followed by the fitness calculation. The procedure of fitness calculation is carried
out for employed bees as well as the onlooker bees. The algorithm is terminated only
when the conditional criteria are met. By using ABC algorithm the intensity
64
parameters are adjusted and improved output image is obtained. The simple ABC
algorithm is explained below.
3.4.2
n-dimensional function f(x), where xi is the i-th candidate solution (adapted from
[179]).
65
N population size
Initialize the positive integer L, which is the stagnation limit
Initialize the forager population size P f < N
Initiallize the onlooker population size Po = N - Pf
Initialize a random population of foragers xi for i 1, Pf
Initialize the forager trial counters T (xi) = 0 for i 1, Pf
while not (termination criterion)
Forager Bees
For each forager xi, i 1, Pf
k random integer
s random integer
1, N
1, n
such that k i
r U [-1,1]
vi (s) xi (s) + r (xi (s) - xk (s))
If f(vi) is better than f(xi) then
xi vi
T(xi) 0
else
T(xi) T(xi) + 1
End if
Next forager
Onlooker Bees
For each onlooker vi , i 1, Po
66
N population size
Initialize the positive integer L, which is the stagnation limit
Initialize the forager population size P f < N
Initiallize the onlooker population size Po = N - Pf
Initialize a random population of foragers xi for i 1, Pf
Initialize the forager trial counters T (xi) = 0 for i 1, Pf
while not (termination criterion)
Forager Bees
For each forager xi, i 1, Pf
k random integer
s random integer
1, N
1, n
such that k i
r U [-1,1]
vi (s) xi (s) + r (xi (s) - xk (s))
If f(vi) is better than f(xi) then
xi vi
T(xi) 0
else
T(xi) T(xi) + 1
End if
Next forager
Onlooker Bees
For each onlooker vi , i 1, Po
1, Pf
1, n
r U [-1,1]
vi (s) xj (s) + r (xj (s) - xk (s))
If f(vi) is better than f(xj) then
xj vi
T(xj) 0
else
T(xj) T(xj) + 1
End if
Next onlooker
such that k j
67
Scout Bees
For each forager xi, i 1, Pf
If T(xi) > L then
xi randomly-generated individual
T(xi) 0
End if
Next forager
Next generation
Xi
Therefore
) is
a D-dimensional vector. After finding the initial food source calculate the Fitness
function (F) for a new food source (new solution). The Fitness function (F) can be
calculated for getting the maximum Fitness function (F). Therefore the maximum
Fitness function (F) can be obtained by Equation (3.11).
The objective function or Fitness function (F) for the MEMS design is
represented by the Equation (3.11),
F D
(3.11)
where, DA is the Die Area. The Die Area (DA) can be evaluated by the following
Equation (3.12),
D ( x m 2 Z 2 ( Z 3 Z 1 ) 2 H 2 ) y m
(3.12)
68
Die Area (DA) value can be between 90000 to 160000m2 and it can be
relaxed up to 240000 m2. The maximum fitness value is considered as the best food
source and by keeping this Fitness function (F) as initial stage we start our searching
process by using employed bees, onlooker bees, and scout. The initial stage of
Fitness function (F) is calculated.
Start
Initialize the population
Evaluate the population of new solution xi
Calculate fitness
Calculate fitness
Yes
Are termination criteria satisfied?
No
Send scout for best solutions
No
End
69
Employed Bees
The employed bee searches the neighborhood of its current food source
(solution) to find out a new food source (new solution) using the Equation (3.13),
C f X f f X i X l
(3.13)
l
where,
f
next value of
and
l
are the arbitrary selected symbols. Here
l
or
is the very
f
(solution).
solution (food source), the quantity of it will be determined and a greedy choice
process will be presented. If the quality of a new food source (solution) is improved
than the existing position, the employed bee neglects that position and moves
towards a new solution (food source), or else the fitness of a new solution (food
Xi
source) is equal or improved than that of
Xi
place of
3.4.4
70
pi
obtained from all of the employed bees. The probability
F
n
F
f 1
(3.14)
Xi
F
where
. After
choosing a food source (solution) the onlooker bees generates a new food source
using Equation (3.13). After generating the new food source a greedy selection will
be applied same as that we applied in the case of employed bees. If a solution
obtained a food source cannot be improved by these trials then that food source will
be considered as abandoned and the employed bee related to that solution (food
source) becomes a scout. The scout randomly produces new food source and
therefore in our proposed method before the scout process we obtain our improved
better food source (solution).
3.5
Z1, Z2, Z3, yk and Fitness function (F) or Die Area (DA). 1000 iterations were used to
obtain the optimal design.
The optimization process has minimized the Die Area (DA), represented
by the objective or Fitness function (F), by satisfying the design criteria. The best
performing design was saved for each successive starting population to converge on
the optimum values.
71
The results had been displayed for the every 100 iterations up to 500
iterations and the optimum values obtained by the ABC algorithm have also been
described in the Table 3.1 to Table 3.5. After 1000 iterations have been computed,
the best performing optimized design parameters are given as the output. The Fitness
function (F) values using ABC, GA are compared and optimized Fitness function (F)
was reported.
After 100 iterations using the Artificial Bee Colony algorithm method the
top 5 ranks are displayed in the Table 3.1. Among these five rank values using rank
selection method Rank 1 which produce the optimized parameter values
Z1=
2.814 x 10-05, Z2= 5.817 x 10-05, Z3= 4.746 x 10-04, yk= 1.009 x 10-04,
F= 111739.81521 m2. The minimized objective function or Fitness function (F)
at 100th iteration is F= 111739.81521 m2. Figure 3.5 represents the convergence of
the objective function or Fitness function (F) at 100th iteration.
Z1
Z2
Z3
yk
Rank 1
2.814e-05
5.817e-05
4.746e-04
1.009e-04
111739.81521
Rank 2
2.592e-05
6.620e-05
4.654e-04
1.004e-04
111820.19521
Rank 3
2.592e-05
6.620e-05
4.654e-04
1.004e-04
111913.99414
Rank 4
2.592e-05
6.620e-05
4.654e-04
1.004e-04
111913.99414
Rank 5
2.592e-05
6.620e-05
4.654e-04
1.004e-04
111913.99414
72
After 200 iterations using the Artificial Bee Colony algorithm method the
top 5 ranks are displayed in the Table 3.2. Among these five rank values using rank
selection method Rank 1 which produce the optimized parameter values
Z1=
2.100 x 10-05, Z2= 5.446 x 10-05, Z3= 4.897 x 10-04, yk= 1.010 x 10-04,
F= 111913.994141 m2. The minimized objective function or Fitness function
(F) at 200th iteration is F= 111913.99414 m2. Figure 3.6 represents the
convergence of the objective function or Fitness function (F) at 200th iteration.
Table 3.2 Five Optimal Values Obtained in the 200th Iteration
by ABC Algorithm
Iteratio
n
Z1
Z2
Z3
yk
73
Rank 1
2.100e-05
5.446e-05
4.897e-04
1.010e-04
111913.99414
Rank 2
2.100e-05
5.446e-05
4.897e-04
1.010e-04
111979.77470
Rank 3
2.100e-05
5.446e-05
4.897e-04
1.010e-04
111979.77470
Rank 4
2.100e-05
5.446e-05
4.897e-04
1.010e-04
111979.77470
Rank 5
2.315e-05
6.079e-05
4.800e-04
1.007e-04
112358.53887
Z1=
2.100 x 10-05, Z2= 5.446 x 10-05, Z3= 4.897 x 10-04, yk= 1.010 x 10-04,
F= 110409.08721 m2. The minimized objective function or Fitness function (F)
at 300th iteration is F= 110409.08721 m2. Figure 3.7 represents the convergence of
the objective function or Fitness function (F) at 300th iteration.
Table 3.3 Five Optimal Values Obtained in the 300th Iteration by ABC Algorithm
Iteratio
Z1
Z2
Z3
yk
74
n
Rank 1
2.100e-05
5.446e-05
4.897e-04
1.010e-04
110409.08721
Rank 2
2.100e-05
5.446e-05
4.897e-04
1.010e-04
110409.08721
Rank 3
2.150e-05
5.476e-05
4.832e-04
1.052e-04
110809.08721
Rank 4
2.172e-05
5.482e-05
4.869e-04
1.066e-04
110909.08721
Rank 5
2.315e-05
6.079e-05
4.800e-04
1.007e-04
111158.53887
Z1=
2.283 x 10-05, Z2= 5.174 x 10-05, Z3= 4.805 x 10-04, yk= 1.019 x 10-04,
F= 111418.71974 m2. The minimized objective function or Fitness function (F)
at 400th iteration is F= 111418.71974 m2. Figure 3.8 represents the convergence of
the objective function or Fitness function (F) at 400th iteration.
Table 3.4 Five Optimal Values Obtained in the 400th Iteration
by ABC Algorithm
75
Iteratio
n
Z1
Z2
Z3
yk
Rank 1
2.283e-05
5.174e-05
4.805e-04
1.019e-04
111418.71974
Rank 2
2.500e-05
5.777e-05
4.727e-04
1.017e-04
111751.39832
Rank 3
2.500e-05
5.777e-05
4.727e-04
1.017e-04
111751.39832
Rank 4
2.500e-05
5.777e-05
4.727e-04
1.017e-04
111751.39832
Rank 5
3.786e-05
5.071e-05
4.866e-04
1.003e-04
112075.01720
Z1=
2.602 x 10-05, Z2= 5.274 x 10-05, Z3= 4.625 x 10-04, yk= 1.019 x 10-04,
F= 115826.60541 m2. The minimized objective function or Fitness function (F)
at 500th iteration is F= 115826.60541 m2. Figure 3.9 represents the convergence
of the objective function or Fitness function (F) at 500th iteration.
76
Z1
Z2
Z3
yk
Rank 1
2.602e-05
5.274e-05
4.625e-04
1.019e-04
115826.60541
Rank 2
2.602e-05
5.274e-05
4.625e-04
1.019e-04
115826.60541
Rank 3
2.602e-05
5.274e-05
4.625e-04
1.019e-04
115826.60541
Rank 4
2.500e-05
5.777e-05
4.732e-04
1.017e-04
120826.60541
Rank 5
3.486e-05
5.071e-05
4.825e-04
1.003e-04
128075.01720
77
Optimum Design
Property Value
Z1
Z2
Z3
yk
Fitness function(F)
110969.1602 m2
The Table 3.7 shows the comparison of the fitness value between the
ABC algorithm method and the existing method using the Genetic Algorithm. The
Fitness function (F) or Die Area (DA) was minimized from 112297.95 m2 to
110969.16 m2 using the ABC method. The fitness value proved to have improved
than the existing (GA) method.
Table 3.7 Comparison of Fitness Function (F) Value using GA
and ABC Method
Methods
GA method
112297.9516
ABC method
110969.1602
78
111000
110500
110000
GA
ABC
Algorithms
Figure 3.10
3.6
SUMMARY
In this chapter, we have proposed a system to perform the optimization of
the design parameters in the MEMS accelerometer. For this, we have employed
Artificial Bee Colony (ABC) algorithm which provides an efficient optimization
technique. From the simulation study, it is observed that Artificial Bee Colony
(ABC) algorithm method has delivered better results in terms of the fitness values
when compared to other optimization techniques like Genetic Algorithm and also
better optimization.
This algorithm will overcome the issues of GA, and the ABC based
design parameter optimization technique will helpful to design MEMS
accelerometer architecture. The fitness is based on the parameter Die Area (DA)
with a specified range.
79
CHAPTER 4
OPTIMIZATION OF PARAMETERS FOR
MEMS ACCELEROMETER WITH COMBINATION
OF ARTIFICIAL BEE COLONY (ABC) ALGORITHM
AND PARTICLE SWARM OPTIMIZATION (PSO)
4.1
INTRODUCTION
The optimization of parameters of a MEMS accelerometer was discussed
in this chapter. Artificial Bee Colony (ABC) optimization algorithm and Particle
Swarm Optimization (PSO) algorithm used to optimize parameters L1, L2, L3, bf and
Fitness function (F) or Die Area (DA) values of MEMS accelerometer. ABC
performs primary optimization and PSO does the optimization of the Fitness
solution resulting from the execution of the ABC algorithm. Die Area (DA) along
with the design parameter force (N) is minimized. This optimization process is
carried out in MATLAB 7.12 environment, simulation results are discussed and
optimized parameters are reported. The Fitness function (F) value using ABC with
PSO are compared with existing ABC alone.
4.2
on are some of the parameters involved in the design of MEMS accelerometer. But
the Die Area (DA) and Force form the main parameters of the design. The
80
Figure 4.1 represents the various steps involved in the optimization of parameters of
MEMS accelerometer.
Start
Initialization of Particle
PSO Algorithm
Evaluate Fitness
Stop
Figure 4.1
4.3
accomplished through the selection of various design parameters like Beam length,
Beam width, Beam depth, Beam mass, proof mass etc. The process of MEMS
designing will start with the approaching of the system and selection of the concept
81
domain (e.g. accelerometer design) by the MEMS designer because only then the
design specifications can be formulated. The designer then inputs the design
specifications for these various parameters. The parameters and its specifications can
be represented as follows:
Beam Length
Beam Width
Beam Depth
Beam Mass
M {aM, bM}
Proof Mass
f { af, bf }
(4.1)
G1 G2 G3 G j Q
(4.2)
Where P=Q=1.8m. The MEMS accelerometer diagram indicates all the
parameters that can be utilized to yield an optimal design of MEMS. A folded beam
structure is available for use and the parameter values are described as in
Equation (4.3) to Equation (4.6),
Lj R
(4.3)
bM T
(4.4)
82
a M L j 2W1
(4.5)
a f W j 2 L2 2W1 2W3
(4.6)
The value of R= 150m and the value of T= 100m. The other
parameters like L1, L2, L3 and bf assigned to B, D, E, F using Equation (4.7) to
Equation (4.10),
L1 B
(4.7)
L2 D
(4.8)
L3 E
(4.9)
bf F
(4.10)
Where, 20m B 500m, 20m D 100m, 100m E
500m, 100m F 500m [11]. Optimization of the parameters is done with
both the ABC algorithm and PSO algorithm.
83
W2
W3
L2
Wj
Lj
Anchor
W2
L2
bM
L1
L3
aM
Proof Mass
bf
af
Anchor
motivated by the natural behavior of honey bees in finding their best food resources.
The colony of artificial bees in the ABC algorithm has three clusters of bees,
namely, employed bees, onlookers and scouts. The initialization procedure consists
of random generation of a set of food source positions as well as the assignment of
values to the control parameters of the algorithm.
The quantity of nectar obtained from the food source denotes the quality
of solution embodied in that food source. Hence, the nectar amounts of the food
sources available at the initial positions are determined. A bee that remains in the
dance area to gather the information regarding food sources is termed as an
84
onlooker. A bee that visits the food source is called as an employed bee. A scout bee
is the one that performs random search.
The bees in the ABC model aim at discovering the best solution. The
location of a food source indicates a possible solution to the optimization problem
and the amount of nectar specifies the quality (fitness) of the solution related to the
food source. On delivering the information regarding the food source to the
onlookers, the employed bee would visit the food source area that was visited by her
previously in the past cycle using the food source information residing in her
memory and then, continues to select a new food source that lies in the
neighborhood of the previously visited food source through visual information and
evaluates its nectar amount.
The same process is repeated by the employed bee in the second stage
after giving food source information resulting from first stage to onlookers, except
that the new food source will now lie in the neighborhood of the food source visited
at the first stage. During the third stage, an onlooker uses the nectar information
given by the employed bees in the dance area to choose the food source area. If the
source gets discarded, the working bee becomes a survey bee and begins to hunt a
novel starting place in the region seal of the colony.
The amount of working bees or the observer bees is corresponding to the
number of results in the inhabitants. In the first cycle, ABC generates an arbitrarily
spread early inhabitants of results. Sometimes ago compute gets finished, the
inhabitants of the result will be subjected to frequent cycles of the hunt procedure
handled by the working bees, the observer bees and the survey bees. A working bee
is the one that modifies the solution in its memory with the help of area information
and then, checks the quantity of nectar in the new solution.
The employed bee would remember the new position, if the nectar
quantity of the present food source is larger than in the previous food source and
discards information regarding the previous food source. Else if the nectar quantity
in the present food source is smaller than in the previous food source, the
85
information of the previous food source is retained in the memory of the bee. On
completion of the search process, the employed bees share the information related to
the position of the food source and the nectar amount of the food source to the
onlookers.
The onlookers would then make an estimate of the information obtained
from the employed bees and chooses a food source that has a probability identical to
its quantity. If the solution achieved is satisfactory, memorize the solution and end
up the process or else, continue the process until the new solution attains the suitable
criteria. The working procedure of the ABC algorithm given in Figure 4.3.
Figure 4.3 reveals the complete process of ABC algorithm. Initialization
of the population forms the first stage of ABC algorithm, which is then followed by
the fitness calculation. The fitness calculation process is done for both the employed
bees and the onlooker bees. The algorithm gets ended only if the conditional criteria
are satisfied. Through the usage of ABC algorithm, the intensity parameters are
adjusted and an enhanced output image is obtained. The proposed ABC algorithm is
explained below.
86
Start
Initialization
Evaluate new solution(s)
Calculate fitness
Yes
No
Send scout for best
solutions
Are termination criteria satisfied?
Yes
Memorize the best solution
Stop
No
87
Ri
Therefore,
) is
a D-dimensional vector. After finding the initial food source, calculate the Fitness
function (F) for a new food source (new solution). The Fitness function (F) is
calculated to yield the maximum Fitness function (F). Therefore, the maximum
Fitness function (F) can be obtained by Equation (4.11).
The objective function or Fitness function (F) for the MEMS design is
represented by Equation (4.11),
F D N
(4.11)
N
where, DA is the Die Area and
N ma
(4.13)
m
where,
is the acceleration.
Die Area (DA) value can range between 90,000 to 160,000m2 and it can
be relaxed up to 240,000 m2 [11].
The food source with the maximum fitness value is taken to be the best
food source and by keeping this Fitness function (F) as initial stage, searching
process that employs the employed bees, onlooker bees and scout is initiated. The
initial stage of Fitness function (F) is calculated.
88
The employed bee searches the neighborhood of its current food source
(solution) to find out a new food source (new solution) using Equation (4.14),
K f R f f R i R l
(4.14)
l
where,
f
of
and
l
are the randomly chosen symbols. Here,
l
or
is a
random number between -1 and +1. After creating the new solution (food source),
the quantity of it will be determined and a greedy choice process will be presented.
If the quality of a new food source (solution) is better than the present position, the
employed bee neglects that position and moves towards new solution (food source),
or else the fitness of a new solution (food source) is equal or improved than that of
Ri
Ri
. Therefore, the new solution takes the place of
a new solution.
The onlooker bees estimate the knowledge gathered from all of the
Si
employed bees to make a selection of the food source. The probability
of
F (t )
n
F (t )
f 1
(4.15)
89
F (t )
where,
Ri
is the fitness value of the solution (food source)
. After choosing a
food source (solution), the onlooker bees generate a new food source using Equation
(4.14). Once the new food source is selected, a greedy selection will be applied in a
similar way as that applied to employed bees. A food source is believed to be
abandoned if a solution obtained a food source that cannot be improved by these
trials and the employed bee associated with that solution (food source) will now turn
out to be a scout. The scout creates a new food source in a random manner and
hence in our proposed method, improved better food source (solution) is obtained
prior to scout process. In this way, various optimal design parameters for the
accelerometer are obtained for the purpose of improving the design process of the
MEMS accelerometer. The obtained fitness values for the design parameters are
optimized again using the PSO algorithm to get a better optimized solution.
4.5
using PSO algorithm. This part explains about introduction to swarm intelligence
and PSO, a simple PSO algorithm, various steps involved in the PSO.
4.5.1
90
4.5.2
Introduction to PSO
An Evolutionary computational technique based on the movement and
intelligence of swarms looking for the most fertile feeding location. It is a Simple
algorithm, easy to implement and few parameters to adjust mainly velocity.
A swarm is an apparently disorganized collection (population) of moving individuals
that tend to cluster together while each individual seems to be moving in a random
direction. It uses a number of particles that constitutes a swarm moving around in
the search space looking for the best solution. Each particle is treated as a point in a
D-dimensional space which adjusts its flying according to its own flying
experience as well as the flying experience of other particles. Each particle keeps
track of its coordinates in the problem space which are associated with the best
solution (fitness) that has achieved so far. This value is called pbest. Another best
value that is tracked by the PSO best value obtained so far by any particle in the
neighbors of the particle is called gbest. The PSO concept consists of changing the
velocity of each particle toward its pbest and the gbest position at each time step.
Each particle tries to modify its current position and velocity according to the
distance between its current position and the pbest, and the distance between its
current position and gbest [166]. The velocity and position update can be done by
91
(4.17)
4.5.3
92
1, N
Hi nearest neighbors of xi
hi arg minx
f (x)
: x Hi
b i - xi
+ 2
hi - xi
max
/i
End if
xi xi + vi
bi arg min f (xi), f (bi)
Next individual
Next generation
4.5.4
Steps in PSO
The general steps involved in the Particle Swarm Optimization are
explained as follows. In step 1 initialize the swarms from the solution space. In step
2 evaluate fitness of individual particles. If termination criteria or condition is met
obtain the best solution otherwise, in step 3 modify pbest, gbest and velocity. In
step 4 move each particle to a new position. In step 5 Go to step 2, and repeat until
convergence or stopping condition is satisfied.
93
4.6
The application of ABC and PSO algorithms has produced optimal values for L1, L2,
L3 and bf. A total of thousand iterations were used to accomplish the optimal design.
The optimization process has minimized the Die Area (DA) along with force
parameter, represented by the objective or Fitness function (F), through the
satisfaction of design criteria. The Tables from 4.1 to 4.10 depicts the optimal values
obtained for 50th, 100th, 150th, 200th, 250th, 300th, 350th, 400th, 450th and 500th iteration
respectively. Five optimal values for each of the 50 iterations are found out and
tabulated. The corresponding graphical representations are shown from
Figures 4.4 to 4.13 with the minimized objective function. The Fitness function (F)
values using ABC with PSO and ABC alone are compared and the optimized Die
Area (DA) was reported.
After 50 iterations using the ABC with PSO algorithm method the top 5
ranks are displayed in the Table 4.1. Among these five rank values using rank
selection method Rank 1 which produce the optimized parameter values
L1= 2.180 x 10-05, L2= 4.695 x 10-05, L3= 4.900 x 10-04, bf= 1.022 x 10-04,
F= 115777.9014 m2. The minimized objective function or Fitness function (F)
at 50th iteration is F= 115777.9014 m2. Figure 4.4 represents the convergence of
the objective function or Fitness function (F) at 50th iteration.
Table 4.1 Five Optimal Values Obtained in the 50th Iteration
after Optimization
Iteratio
n
L1
L2
L3
bf
Rank 1
2.180e-05
4.695e-05
4.900e-04
1.022e-04
115777.9014
Rank 2
2.180e-05
4.695e-05
4.900e-04
1.022e-04
116188.00965
Rank 3
2.986e-05
4.830e-05
4.844e-04
1.027e-04
116578.33164
Rank 4
2.491e-05
6.257e-05
4.642e-04
1.023e-04
116588.48896
Rank 5
2.491e-05
6.257e-05
4.642e-04
1.023e-04
116588.48896
94
L1
L2
L3
bf
Rank 1
2.283e-05
5.174e-05
4.805e-04
1.019e-04
115826.6054
Rank 2
2.500e-05
5.777e-05
4.727e-04
1.017e-04
119751.3983
Rank 3
2.500e-05
5.777e-05
4.727e-04
1.017e-04
121751.3983
Rank 4
2.500e-05
5.777e-05
4.727e-04
1.017e-04
125751.3983
Rank 5
3.786e-05
5.071e-05
4.866e-04
1.003e-04
127075.0172
95
L1
L2
L3
bf
Rank 1
2.679e-05
4.533e-05
4.980e-04
1.007e-04
110743.1452
Rank 2
2.679e-05
4.533e-05
4.980e-04
1.007e-04
118053.07598
Rank 3
2.679e-05
4.533e-05
4.980e-04
1.007e-04
121053.07598
Rank 4
2.679e-05
4.533e-05
4.980e-04
1.007e-04
185053.07598
Rank 5
3.070e-05
5.371e-05
4.838e-04
1.005e-04
224669.52840
96
L1
L2
L3
bf
Rank 1
3.015e-05
4.464e-05
4.925e-04
1.020e-04
110706.21331
Rank 2
3.015e-05
4.464e-05
4.925e-04
1.020e-04
110706.21331
Rank 3
3.863e-05
4.527e-05
4.914e-04
1.015e-04
110706.21331
Rank 4
3.863e-05
4.527e-05
4.914e-04
1.015e-04
110706.21331
Rank 5
3.863e-05
4.527e-05
4.914e-04
1.015e-04
110706.21331
97
L1
L2
L3
bf
Rank 1
2.367e-05
4.889e-05
4.950e-04
1.004e-04
110457.56342
Rank 2
2.206e-05
4.814e-05
4.817e-04
1.036e-04
110592.78178
Rank 3
2.206e-05
4.814e-05
4.817e-04
1.036e-04
110692.78178
Rank 4
2.206e-05
4.814e-05
4.817e-04
1.036e-04
110792.78178
Rank 5
2.206e-05
4.814e-05
4.817e-04
1.036e-04
110892.78178
98
L1
L2
L3
bf
Rank 1
2.904e-05
4.905e-05
4.901e-04
1.001e-04
110119.3144
Rank 2
2.422e-05
4.210e-05
4.948e-04
1.022e-04
110119.3144
Rank 3
2.412e-05
4.954e-05
4.922e-04
1.014e-04
110119.3144
Rank 4
2.414e-05
3.907e-05
4.975e-04
1.045e-04
110119.3144
Rank 5
3.009e-05
5.555e-05
4.876e-04
1.006e-04
110119.3144
99
L1
L2
L3
bf
Rank 1
2.703e-05
5.563e-05
4.763e-04
1.016e-04
110602.9662
Rank 2
3.989e-05
4.607e-05
4.961e-04
1.005e-04
112555.50022
Rank 3
2.019e-05
6.605e-05
4.604e-04
1.030e-04
116171.31011
Rank 4
2.019e-05
6.605e-05
4.604e-04
1.030e-04
118171.31011
Rank 5
2.019e-05
6.605e-05
4.604e-04
1.030e-04
124171.31011
100
L1
L2
L3
bf
Rank 1
2.151e-05
4.177e-05
4.939e-04
1.037e-04
109618.2632
Rank 2
2.151e-05
4.177e-05
4.939e-04
1.037e-04
109618.2632
Rank 3
2.044e-05
7.413e-05
4.514e-04
1.019e-04
109618.2632
Rank 4
2.044e-05
7.413e-05
4.514e-04
1.019e-04
109618.2632
Rank 5
2.315e-05
3.750e-05
4.885e-04
1.066e-04
109618.2632
101
L1
L2
L3
bf
Rank 1
2.754e-05
4.534e-05
4.919e-04
1.014e-04
110165.74376
Rank 2
2.754e-05
4.534e-05
4.919e-04
1.014e-04
111053.33436
Rank 3
2.754e-05
4.534e-05
4.919e-04
1.014e-04
111053.33436
Rank 4
2.169e-05
5.984e-05
4.793e-04
1.004e-04
111468.25048
Rank 5
3.417e-05
5.371e-05
4.862e-04
1.001e-04
112047.32099
102
L1
L2
L3
bf
Rank 1
2.754e-05
4.534e-05
4.919e-04
1.014e-04
110830.8728
Rank 2
2.754e-05
4.534e-05
4.919e-04
1.014e-04
111053.33436
Rank 3
2.754e-05
4.534e-05
4.919e-04
1.014e-04
111053.33436
Rank 4
2.169e-05
5.984e-05
4.793e-04
1.004e-04
111468.25048
Rank 5
3.417e-05
5.371e-05
4.862e-04
1.001e-04
112047.32099
103
ABC method
110969.1602
110409.08701
104
Comparison of Fitness function (F) value using ABC and ABC with PSO Algorithms
111200
111000
110800
Fitness function (F) value in m2
110600
110400
110200
110000
ABC
Algorithms
Figure 4.14
4.7
SUMMARY
This chapter deals with a system that optimizes the design parameters of
105
CHAPTER 5
COMPARISON OF OPTIMAL PARAMETER VALUES
OF MEMS ACCELEROMETER USING THREE DIFFERENT
OPTIMIZATION TECHNIQUES
5.1
INTRODUCTION
In this chapter the optimized Beam Length values ( L1, L2 and L3), Proof
Mass(ym) and Fitness function (F) values of MEMS Accelerometer using Genetic
Algorithm, Artificial Bee Colony Algorithm, Artificial Bee Colony (ABC) with
Particle Swarm Optimization (PSO) Algorithm methods are compared. Based on the
comparison of the values optimized parameters of a MEMS accelerometer were
reported.
5.2
Mass (ym) and Fitness function (F) values using GA, ABC, and ABC with PSO
Algorithm methods.
106
Table 5.1
Comparison of Beam Length Values (L1, L2, L3), Proof Mass (ym)
and Fitness Function (F) Value using the GA, ABC and ABC with
PSO Algorithm Methods
Optimized
Design Value
5.3
Property Value
Using GA
Property Value
Using ABC
Property Value
Using ABC &PSO
L1
3.085 x 10-04
= 308.5 m
2.364 x 10-05
= 23.64 m
2.114 x 10-05
= 21.14 m
L2
6.019 x 10-05
= 60.19 m
5.356 x 10-05
= 53.56 m
4.995 x 10-05
= 49.95 m
L3
4.107 x 10-04
= 410.7 m
4.938 x 10-04
= 493.8 m
4.854 x 10-04
= 485.4 m
ym
1.019 x 10-04
= 101.9 m
1.015 x 10-04
= 101.5 m
1.013 x 10-04
= 101.3 m
112297.95163 m2 110969.1602 m2
110409.08701 m2
L1= 3.085 x 10-04 (GA), L1= 2.364 x 10-05(ABC), L1= 2.114 x10-05(ABC with PSO).
Here the Beam Length L1 was optimized or minimized from 308.5 m (GA) to
21.14 m (ABC with PSO). So the optimized Beam Length (L1) reported from
the Table 5.1 was L1= 21.14 m. This is the optimized Beam Length ( L1) value
obtained by using ABC with PSO method.
The Figure 5.1 shows the comparison of Beam Length (L1) values
graphically. The different types of algorithms taken in X-axis and Beam Length (L1)
in m taken in Y-axis. It proves that ABC with PSO Algorithm optimized the
Beam Length (L1) in a better way compared to GA, ABC methods.
107
Comparison of Beam Length (L1) Value using GA, ABC, ABC with PSO Algorithms
0
0
0
0
Beam Length (L1) in m 0
0
0
0
GA
ABC
ABC &PSO
Algorithms
Figure 5.1 Comparison of Beam Length (L1) Value using the GA, ABC and
ABC with PSO Algorithm methods
The Beam Length L2 value obtained using GA, ABC, ABC with PSO are
L2= 6.019 x 10-05 (GA), L2= 5.356 x 10-05 (ABC), L2= 4.995 x 10-05 (ABC with
PSO). Here the Beam Length L2 was optimized or minimized from 60.19 m
(GA) to 49.95 m (ABC with PSO). So the optimized Beam Length (L2)
reported from the Table 5.1 was L2= 49.95 m. This is the optimized Beam
Length (L2) value obtained by using ABC with PSO method.
The Figure 5.2 shows the comparison of Beam Length (L2) values
graphically. The different types of algorithms taken in X-axis and Beam Length (L2)
in m taken in Y-axis. It proves that ABC with PSO Algorithm optimized the
Beam Length (L2) in a better way compared to GA, ABC methods.
108
Comparison of Beam Length (L2) Value using GA, ABC, ABC with PSO Algorithms
0
0
0
0
Beam Length (L2) in m
0
0
0
0
GA
ABC
ABC &PSO
Algorithms
Figure 5.2 Comparison of Beam Length (L2) Value using the GA, ABC and
ABC with PSO Algorithm Methods
The Beam Length L3 value obtained using GA, ABC, ABC with PSO are
L3= 4.107 x 10-04 (GA), L3= 4.938 x 10-04 (ABC), L3= 4.854 x 10-04 (ABC with
PSO). Here the Beam Length L3 was not optimized in a better way and it
increases from 410.7 m (GA) to 485.4 m (ABC with PSO). Even though it is
not optimized but it used to optimize the Fitness function (F) in ABC with PSO
method. So the Beam Length (L3) reported from the Table 5.1
was L3= 485.4 m. This is the Beam Length (L3) value obtained by using
ABC with PSO method.
The Figure 5.3 shows the comparison of Beam Length (L3) values
graphically. The different types of algorithms taken in X-axis and Beam Length (L3)
in m taken in Y-axis.
109
Comparison of Beam Length (L3) Value using GA, ABC, ABC with PSO Algorithms
0
0
0
Beam Length (L3) in m
0
0
0
0
GA
ABC
ABC &PSO
Algorithms
Figure 5.3 Comparison of Beam Length (L3) Value using the GA, ABC and
ABC with PSO Algorithm Methods
5.4
ym = 1.019 x 10-04 (GA), ym = 1.015 x 10-04 (ABC), ym = 1.013 x 10-04 (ABC with
PSO). Here the Proof Mass (ym) was optimized or minimized from 101.9 m
(GA) to 101.3 m (ABC with PSO). So the optimized Proof Mass (ym) reported
from the Table 5.1 was ym = 101.3 m. This is the optimized Proof Mass (ym)
value obtained by using ABC with PSO method. The Figure 5.4 shows the
comparison of Proof Mass (ym) values graphically. The different types of algorithms
taken in X-axis and Proof Mass (ym) in m taken in Y-axis. It proves that ABC
with PSO Algorithm optimized the Proof Mass (ym) in a better way compared to
GA, ABC methods.
110
Comparison of Proof Mass (ym) Value using GA, ABC, ABC with PSO Algorithms
0
0
0
Proof Mass (ym) in m
0
0
0
0
GA
ABC
ABC &PSO
Algorithms
Figure 5.4 Comparison of Proof Mass (ym) Value using the GA, ABC and ABC
with PSO Algorithm Methods
5.5
111
Comparison of Fitness function (F) Value using GA,ABC,ABC with PSO Algorithms
112500
112000
111500
111000
Fitness function (F) in m2 110500
110000
109500
109000
GA
ABC
Algorithms
Figure 5.5 Comparison of Fitness Function (F) Value using the GA, ABC, and
ABC with PSO Algorithm Methods
5.6
SUMMARY
The optimized parameter of a MEMS accelerometer was compared.
112
CHAPTER 6
CONCLUSION
6.1
INTRODUCTION
The present work deals with optimization of parameters of MEMS
= 493.8 m, ym = 1.015 x 10-04 =101.5 m and Fitness function (F) or Die Area
113
To reduce the Fitness function (F) or Die Area (DA) still further a new
algorithm (combination of ABC with PSO) has been introduced. Here force was
introduced along with Die Area (DA) as a Fitness function (F).
The optimized parameter values L1 = 2.114 x 10-05 = 21.14 m, L2 =
4.995 x 10-05 = 49.95 m, L3 = 4.854 x 10-04 = 485.4 m, ym = 1.013 x 10-04 =101.3
m and Fitness function (F) or Die Area (DA) = 110409.09 m2 are obtained
from simulation. By comparing these parameters L1, L2, L3, ym and Fitness
function (F) or Die Area (DA) values with the parameter values obtained using
Artificial Bee Colony (ABC) algorithm, Genetic Algorithm (GA), we conclude
that the parameter values obtained using Artificial Bee Colony (ABC)
algorithm with Particle Swarm Optimization (PSO) algorithm method is better
than that the parameter values obtained using Genetic Algorithm (GA) and
Artificial Bee Colony (ABC) algorithm method.
6.2
RESEARCH CONTRIBUTION
Based on the simulation results obtained the optimized parameter
114
115
REFERENCES
[1] Zhang, Y., Kamalian, R., Agogino, A.M., and Squin, C.H., Hierarchical
MEMS Synthesis and Optimization, Proc. of SPIE, Bellingham, WA,
5763, pp. 96-106, 2005.
[2] Rubio, W.M., Godoy, P.H.de., and Silva, E.C.N., Design of
Electrothermomechanical MEMS, ABCM Symposium Series in
Mechatronics, 2, pp. 469-476, 2006.
[3] Neukermans, A., and Ramaswami, R., MEMS technology for optical
networking applications, Communications Magazine, IEEE., 39(1),
pp. 62-69, 2001.
[4] Albano, F., et al., Design of an implantable power supply for an intraocular
sensor, using POWER (power optimization for wireless energy
requirements), Journal of Power Sources, 170, pp. 216224, 2007.
[5] Udina, S., et al., A MEMS based compact natural gas analyzer
implementing IEEE-1451.2 and BS-7986 smart sensor standards, The 14th
International Meeting on Chemical Sensors, Tagungsband, pp. 1213-1216,
2012.
[6] Kim, M., Hacker, J. B., Mihailovich, R. E., and DeNatale, J. F., A DC-to40 GHz four-bit RF-MEMS true-time delay network, Microwave and
Wireless Components Letters, IEEE, 11(2), pp. 56-58, 2001.
[7] Cetiner, B. A., et al., Multifunctional reconfigurable MEMS integrated
antennas for adaptive MIMO systems, Communications Magazine,
IEEE, 42(12), pp. 62-70, 2004.
[8] Walter, H., Dudek, R., and Michel, B., Fracture and Fatigue Behavior of
MEMS Related Micro Materials, Proc. of 11th International Conference on
Fracture (ICXI), Italy, pp. 365-384, 2005.
[9] Deng, L., et al., Electrical calibration of spring-mass MEMS capacitive
accelerometers, Proc. of the Conference on Design, Automation and Test in
Europe, Grenoble, France, pp. 571-574, 2013.
[10] Design Optimization of MEMS Comb Accelerometer., Available:
http://www.asee.org/ documents/ zones/zone1/ 2008/ student/
ASEE12008_0050_paper.pdf
116
117
[24] Binitha, S., and Siva Sathya, S., A Survey of Bio inspired Optimization
Algorithms, International Journal of Soft Computing and Engineering,
2(2), pp. 137-151, 2012.
[25] Passino, K.M., Bacterial Foraging Optimization, International Journal
of Swarm Intelligence Research, 1(1), pp. 1-16, 2010.
[26] Thomas, R.M., Survey of Bacterial Foraging Optimization Algorithm,
International Journal of Innovative Science and Modern Engineering, 1(4),
pp. 11-12, 2013.
[27] Vipul Sharma, Pattnaik, S. S., and Tanuj Garg., A Review of Bacterial
Foraging Optimization and Its Applications, IJCA Proc. on National
Conference on Future Aspects of Artificial Intelligence in Industrial
Automation (NCFAAIIA), (1), pp. 9-12, 2012.
[28] Yang, X.S., Firefly Algorithm, Stochastic Test Functions and Design
Optimization, Int.J. Bio-Inspired Computation, 2(2), pp. 78-84, 2010.
[29] Horng, M.H., Lee, Y.X., Lee, M.C., and Liou, R.J., Firefly
metaheuristic algorithm for training the radial basis function network for
data classification and disease diagnosis, In: Parpinelli, R., Lopes, H.S.,
(eds) Theory and New Applications of Swarm Intelligence, pp. 115-132,
2012.
[30] Horng, M.H., Vector quantization using the firefly algorithm for image
compression, Expert Syst. Appl., 39, pp. 1078-1091, 2012.
[31] Banati, H., and Bajaj, M., Firefly based feature selection approach, Int.
J. Computer Science Issues, 8(2), pp. 473-480, 2011.
[32] Azad, S.K., and Azad, S.K., Optimum Design of Structures Using an
Improved Firefly Algorithm, Int. J. Optim.Civil Eng., 1(2), pp. 327-340,
2011.
[33] Gandomi, A. H., Yang, X. S., and Alavi, A. H., Cuckoo search
algorithm: a metaheuristic approach to solve structural optimization
problems, Engineering with Computers, 27, pp. 1-19, 2011.
[34] Basu, B., and Mahanti, G.K., Firefly and artificial bees colony
algorithm for synthesis of scanned and broadside linear array antenna,
Progress In Electromagnetics Research B, 32, pp. 169-190, 2011.
[35] Sayadi, M. K., Ramezanian, R., and Ghaffari-Nasab, N., A discrete
firefly meta-heuristic with local search for make span minimization in
permutation flow shop scheduling problems, Int. J. of Industrial
Engineering Computations, 1, pp. 1-10, 2010.
118
119
[49] Kao, C.C., and Torres, G.L., Applications of PSO to optimal power
systems, Inter. J. of Inno. Comp, Information and control, 8(3A),
pp. 1705-1716, 2012.
[50] Holland, J.H., Genetic algorithms and the optimal allocation of trials,
SIAM J. Comput., 2 (2), pp. 88105, 1973.
[51] Goldberg, D.E., (1989), Genetic algorithms in Search, Optimization,
and Machine learning, Addison-Wesley, Newyork.
[52] Goswami, B., and Mandal, D., A genetic algorithm for the level control
of nulls and side lobes in linear antenna arrays, J.King Saud Univ. Comput.
Inf. Sci., 25, pp. 117-126, 2013.
[53] Applications of Genetic Algorithm., Available: http://en.wikipedia.org/
wiki/List_of_genetic_algorithm_ applications
[54] Applications of Genetic Algorithm., Available: http://www.doc.ic.ac.uk/
~nd/surprise_96/journal/vol1/tcw2/article1.html
[55] Hoang Pham, (2006), Springer Handbook of Engineering Statistics,
pp. 749-773, Springer-Verlag, London.
[56] Applications of Genetic Algorithm., Available: http://neo.lcc.uma.es/
TutorialEA/semEC/cap03/ cap_3.html
[57] Man, K.F., Tang, K.S., and Kwong, S., Genetic Algorithms: Concepts
and Applications, IEEE Trans. on Industrial Electronics, 43(5), pp. 519534, 1996.
[58] Simon, D., Biogeography -Based Optimization, IEEE Transactions on
Evolutionary Computation, 12(6), pp. 702-713, 2008.
[59] Khokhar, B., Parmar, K.P.S., and Dahiya, S., Application of
Biogeography-based Optimization for Economic Dispatch Problems,
International Journal of Computer Applications, 47(13), pp. 25-30, 2012.
[60] Lohokare, M.R., et al., Biogeography based optimization technique for
block based motion estimation in video coding, National Conference on
Computational Instrumentation, CSIO Chandigarh, India, pp. 67-71, 2010.
[61] Gupta, S., Bhuchar, K., and Sandhu, P., Implementing Color Image
Segmentation Using Biogeography Based Optimization, International
120
121
122
123
[94] Li, C., and Zhou, J., Parameters identification of hydraulic turbine
governing system using improved gravitational search algorithm, Energy
Conversion and Management, 52(1), pp. 374381, 2011.
[95] Li, C., Zhou, J., Xiao, J., and Xiao, H., Hydraulic Turbine Governing
System Identification using T- S Fuzzy Model Optimized by Chaotic
Gravitational Search Algorithm, Engineering Applications of Artificial
Intelligence, 260(9), pp. 20732082, 2013.
[96] Parvin, J. R., and Vasanthanayaki, C., Gravitational Search Algorithm
Based Mobile Aggregator Sink Nodes for Energy Efficient Wireless Sensor
Networks, In International Conference on Circuits, Power and Computing
Technologies (ICCPCT-2013), Nagarcoil, India, pp. 10521058, 2013.
[97] Lee, K.S., and Geem, Z.W., A new meta-heuristic algorithm for
continuous engineering optimization: harmony search theory and practice,
Comput. Methods Appl. Mech. Engrg., 194(36-38), pp. 3902-3933, 2005.
[98] Geem, Z., School bus routing using harmony search, In Genetic and
Evolutionary Computation Conference (GECCO), Washington DC, USA,
pp. 1-6, 2005.
[99] Geem, Z.W., Harmony search algorithm for solving sudoku, In:
Apolloni, B., Howlett, R.J., Jain, L., [eds] KES, Part I.LNCN (LNAI),
Springer, Heidelberg, 4692, pp. 371-378, 2007.
[100] Geem, Z., Harmony search algorithm for the optimal design of largescale water distribution network, In Proc. 7th International IWA
Symposium on Systems Analysis and Integrated Assessment in Water
Management, IWA, Washington DC, USA, 2007,(CD-ROM).
[101] Geem, Z., and Hwangbo, H., Application of harmony search to multiobjective optimization for satellite heat pipe design, In Proc. US-Korea
Conference on Science, Technology, and Entrepreneurship , Teaneck, NJ,
USA, pp. 1-3, 2007.
[102] Geem, Z., Lee, K., and Tseng, C., Harmony search for structural
design, In Proc. 2005 conference on Genetic and evolutionary
computation, ACM New York, NY, USA, pp. 651- 652, 2005.
[103] Geem, Z., and Williams, J., Harmony search and ecological
optimization, International Journal of Energy and Environment, 1, pp. 150154, 2007.
[104] Geem, Z. W., Optimal scheduling of multiple dam system using
harmony search algorithm, Lecture Notes in Computer Science, 4507, pp.
316-323, 2007.
124
[105] Geem, Z.W., and Choi, J.Y., Music composition using harmony search
algorithm, Lecture Notes in Computer Science, 4507, pp. 593-600, 2007.
[106] Geem, Z., Lee, K., and Park, Y., Application of harmony search to
vehicle routing, American Journal of Applied Sciences, 2(12), pp. 15521557, 2005.
[107] Al-Betar, M., Khader, A., and Gani, T., A harmony search algorithm for
university course timetabling, In 7th International Conference on the
Practice and Theory of Automated Timetabling (PATAT 2008), Montreal,
Canada, 194(1), pp. 3-31, 2008.
[108] Ryu, S., Duggal, A.S., Heyl, C. N., and Geem, Z. W., Mooring Cost
Optimization via Harmony Search, Proc. 26th ASME International
Conference on Offshore Mechanics and Arctic Engineering (OMAE 2007),
San Diego, CA, USA, 1, pp. 355-362, 2007.
[109] Karahan, H., Gurarslan, G., and Geem, Z.W., Parameter Estimation of
the nonlinear Muskingum flood routing model using a hybrid harmony
search algorithm, J. Hydrol. Eng., 18(3), pp. 352-360, 2012.
[110] Fesanghary, M., Damangir, E., and Soleimani, I., Design optimization of
shell and tube heat exchangers using global sensitivity analysis and
harmony search, Applied Thermal Engineering, 29, pp. 1026-1031, 2009.
[111] Scott Kirkpatricks., Gelatt, C.D., and Vecchi, P.M., Optimization by
simulated annealing, Science, 220(4598), pp. 671680, 1983.
[112] Tian, P., Ma, J., and Zhang, D.M., Application of the simulated
annealing algorithm to the combinatorial optimization problem with
permutation property: An investigation of generation mechanism,
European Journal of Operational Research, 118(1), pp. 81-94, 1999.
[113] Van Laarhoven, J.P., and Aarts, E.H., (1987), Simulated Annealing:
Theory and Applications, 37, pp. 77-98, Kluwer Academic Publishers,
Norwell, USA.
[114] Reynolds, R.G., Introduction to cultural algorithms, Third Annual
Conference on Evolutionary Computing, Antony V.Sebald et al., eds., World
Scientific Press, Singapore, pp. 131-139, 1994.
[115] Cultural Algorithms., Available: http://en.wikipedia.org/wiki/
Cultural_algorithm
[116] Reynolds, R. G., (1999) Cultural Algorithms: Theory and Applications,
New Ideas in Optimization, McGraw-Hill Ltd, Maidenhead, UK,
pp. 367378.
125
126
[129] Iqbal, M.A., Khan, N.K., Mujtaba, H., and Baig, R.A., A Novel
Function Optimization Approach Using Opposition based GA with Gene
Excitation, International Journal of Innovative Computing, Information
and Control, 7(7B), pp. 4263-4276, 2011.
[130] Glover, F., and McMillan, C., The general purpose & scheduling
problem: an integration of MS and AI, Computers and Operations
Research, 13(5), pp. 536-573, 1986.
[131] Glover, F., Tabu Search Part I, ORSA Journal on Computing, 1(3),
pp. 190-206, 1989.
[132] Glover, F., Tabu Search Part II, ORSA Journal on Computing, 2(1),
pp. 4-32, 1990.
[133] Glover, F., Tabu Search: A Tutorial, Interfaces, 20(4), pp. 74-94, 1990.
[134] Rao, R.V., Savsani, V.J., and Vakharia, D.P., Teaching-learning-based
optimization: A novel method for constrained mechanical design
optimization problems, Comput Aided Des., 43, pp. 30315, 2011.
[135] Satapathy, S.C., and Naik, A., Data Clustering Based on TLBO,
Swarm, Evolutionary, and Memetic Computing Lecture Notes in Computer
Science, 7077, pp. 148-156, 2011.
[136] Zou, F., et al., Multiobjective optimization using TLBO algorithm,
Engineering Applications of Artificial Intelligence, 26(4), pp. 1291-1300,
2013.
[137] Nayak, M.R., Nayak, C.K., and Rout, P.K., Application of
Multiobjective TLBO algorithm to optimal power flow problems, 2nd
International Conference on Communication, Computing & Security
[ICCCS], Rourkela, India, 6, pp. 255-264, 2012.
[138] Baghlani, A., and Makiabdi, M.H., TLBO Algorithm for Shape & Size
Optimization of Truss structure with dynamic frequency constraints, IJST
Transactions of Civil Engineering, 37, pp. 409-421, 2013.
[139] Satapathy, S.C., Naik, A., and Parvathi, K., A TBLO based on
orthogonal design for solving global optimization problems, Springer Plus,
2:130, pp. 1-12, 2013.
[140] Ganesh, S.B., and Reddy, S., TLBO for EDP with Valve point loading
effect, International Journal of Education and Applied Research, 4(1),
pp. 9-15, 2014.
127
[141] Kaur, D., and Kaur. R., A design of IIR based digital hearing aids using
TLBO, International Journal of Computer Engineering & Applications,
3(2/3), pp. 182-190, 2013.
[142] Lakshmi Reddy, Y., and Sydulu, M., TLBO Algorithm for
Reconfiguration in Radial Distribution Systems for Loss Reduction,
International Journal of Advanced Engineering and Global Technology,
2(4), pp. 622-626, 2014.
[143] Roy, P.K., Sur, A., and Pradhan, D.K., Optimal Short term hydrothermal scheduling using quasi oppositional teaching learning based
optimization, Engineering Applications of Artificial Intelligence, 26(10),
pp. 2516-2524, 2013.
[144] John K, Sakellaris, Finite Element Analysis of Micro Electro
Mechanical Systems: Towards the integration of MEMS in design and
robust optimal control schemes of smart microstructures, WSEAS
TRANSACTIONS on APPLIED and THEORETICAL MECHANICS, 3,
pp. 114-124, 2008.
[145] Singh, A., Prince, A, A., and Agrawal, V. P., Design Optimization &
Comparison of RF Power Sensors based on MEMS, International Journal
of Recent Trends in Engineering, 1(4), pp. 64-67, 2009.
[146] Chen, X., Cui, W., and Xue, W., Process Modeling and Device-Package
Simulation for Optimization of MEMS Gyroscopes, Journal on ComputerAided Design and Applications, 6(3), pp. 375-386, 2009.
[147] Chandrana, C., et al., Design and Analysis of MEMS Based PVDF
Ultrasonic Transducers for Vascular Imaging, Journal on Sensors, 10, pp.
8740-8750, 2010.
[148] Dugosz, A., Multiobjective evolutionary optimization of MEMS
structures, Computer Assisted Mechanics and Engineering Sciences, 17(1),
pp. 41-50, 2010.
[149] Pathak, R., and Joshi, S., Optimizing reliability modeling of MEMS
devices based on their applications, World Journal of Modeling and
Simulation, 7(2), pp. 139-154, 2011.
[150] Naduvinamani, Sujata.N., Sheeparamatti, B.G., and Kalalbandi,
Sandeep, V., Simulation of Cantilever Based RF-MEMS Switch Using
CoventorWare, World Journal of Science and Technology, 1(8),
pp. 149-153, 2011.
128
[151] Nagpal, P., Mehta, Rangra, K., and Aggarwal, R., Optimization of
Capacitive MEMS Pressure Sensor for RF Telemetry, International Journal
of Scientific & Engineering Research, 2(10), pp. 1-4, 2011.
[152] Jain, S., Chechi, D., and Chawla, P., Performance Study of RF MEMS
Ohmic Series Switch, International Journal of Advanced Research in
Computer Science and Software Engineering, 2(8), pp. 485-488, 2012.
[153] Zhang, Y., Kamalian, R., Agogino, A.M., and Squin, C.H., Design
Synthesis of MicroElectroMechanical Systems Using Genetic Algorithms
with Component-based Genotype Representation, Proc. of the 8th Annual
Conference on Genetic and Evolutionary Computation, ACM Press, New
York, USA, pp. 731-738, 2006.
[154] Jain, A., Greve, D., and Oppenheim, J., A MEMS Transducer for
Ultrasonic Flow Detection, ISARC, Washington, USA, pp. 375386, 2002.
[155] Attoh-Okine, N. O., and Mensah, S., MEMS Application in Pavement
Condition Monitoring-Challenges, ISARC, Washington, USA,
pp. 387391, 2002.
[156] Kamalian, R.H., Agogino, A.M., and Takagi, H., The Role of
Constraints and Human Interaction in Evolving MEMS Designs: Micro
resonator Case Study, In Proc. of ASME Design Engineering Technical
Conferences and Computers and Information in Engineering Conference,
Saltlake city, Utah, USA, pp. 1-9, 2004.
[157] Li, H., and Antonsson, E.K., Evolutionary Techniques in MEMS
Synthesis, 25th Biennial Mechanisms Conf., ASME Design Engineering
Technical Conf., #DETC98/MECH-5840, Atlanta, Georgia, 1998.
[158] Ma, L., and Antonsson, E.K., Automated Mask-layout and Process
Synthesis for MEMS, Proc. of the Modeling and Simulation of
Microsystems Conference, Sandiego, pp. 20-23, 2000.
[159] Li, J., Gao, S., and Liu, Y., Solid-based CAPP for Surface
Micromachined MEMS Devices, Computer-Aided Design, 39(3), pp. 190201, 2007.
[160] Obadat, M., Hosin, L., Bhatti, M. A., and Mclean, B., Full Scale Field
Evaluation of MEMS Based Bi-axial Transducer, Presented at the 82nd
Annual Meeting of the Transportation Research Board, Washington, D.C,
pp. 1-17, 2003.
129
130
LIST OF PUBLICATIONS
In connection with this Thesis
131
I.
JOURNAL
1.
2.
3.
4.
II.
1.
2.
VIATE
132