engineering optimization applications
DESCRIPTION
Engineering Optimization Applications. 柯春旭 義守大學電機系 2010/5/21. Optimization Problem. Design and Fabrication of an Efficient Magnetic Microactuator. I Introduction II Efficient Magnetic Microactuator III Optimal Design of Efficient Magnetic Microactuator - PowerPoint PPT PresentationTRANSCRIPT
Engineering Optimization Applications
柯春旭 義守大學電機系
2010/5/21
Optimization Problem
Design Variables
ModelDesign Specs.
Outputs Objective function
Design and Fabrication of an Efficient Magnetic Microactuator
I Introduction
II Efficient Magnetic Microactuator
III Optimal Design of Efficient Magnetic Microactuator
IV Simulation of Magnetic Microactuator Using the
Macromodel
V Fabrication of Efficient Magnetic Microactuator
Introduction
• Magnetic microactuators have the advantages of
- large force and large deflection
- low driving voltage
• Micromachined microactuators achieve the needs of
- miniaturization,
- mass production and low cost
• Research activities of magnetic microactuator
- design and fabrication
- simulation and optimization
• MEMS applications
micromotors, microrelays, optics, printer.
Study of Magnetic Microactuator
Magnetic microactuator
• a micromachined electromagnet
as flux generator
• a movable microstructure
with the magnetic material
magnetic field interacts with magnetic material to product a force, the microstructure displaced
Permalloy plate
Planar coils
Magnetic core
attractive force
attractive and repulsive force
Study of Magnetic Microactuator
Magnetic flux
0
10
20
30
0 4 8 12n
forc
e (
N)
force vs. coil number n
Planar coils
Basic design of electromagnet is only using planar
coils
• The advantage
- easily fabricated
• The drawbacks
- leakage flux results in low efficiency
- truns n increases, force approaches to a
constant
Study of Magnetic Microactuator
External Flux
Using a non-micromachined external
magnetic field generator [6,10,23,24]
• The advantage
- force effectively increases
• The drawbacks
- filed generator is larger over all device
dimensions
- requiring additional assembling steps
- EM interference in array
Design of magnetic circuit
Chang Liu, et al.
Ranan A. Miller, et al.
Magnetic Circuits
Three types:
1. Spiral Type
• The advantages
- coils are easily fabricated
- accurate line pitch and width
• The drawbacks
- the resistance nonlinearly increases
- internal leakage flux, reduced by using
LIGA or high rMicrorelay, E. Fullin
Micropump, Ahn7 layers
2. Solenoid Type
• The advantages
- the magnetic circuit is easily constructed
- easily achieving the desired shape of
microactuator
• The drawbacks
- not suitable for the fine coil
- non-flat coil contacts raise the resistance,
increasing local high temperature
Microactuator, H. Ren, E. Gerhard
Micropump, SADLER et al.
5 layers
Magnetic Circuits
3. Meander Type
• The advantages
- the magnetic circuit is easily constructed
- the coils are flat
• The drawbacks
- not suitable for the fine coil
- non-flat core contacts increases magnetic reluctance, reducing the efficiency
Microrelay, Marc et al.
5 layers
Magnetic Circuits
Simulation of Magnetic Microactuator
• Detailed knowledge of all of the magneto-structural effects is a prerequisite for effective and efficient design
• Trying the simulation experiments only in hours instead of months, thus shorten the development cycle
• Utilizing the optimization algorithm to achieve the optimal performances of the devices
Magneto-structural coupled problem
(1) To design an efficient magnetic microactuator with the
magnetic circuit.
(2) To optimize the magnetic microactuator for applications.
(3) To develop an efficient macromodeling techniques for dynamic
coupled simulation of magnetic microactuator.
(4) To realize the efficient magnetic microactuator with
micromachining processes.
Objectives
Improved Designpole area can optimally enlarge1. Increasing the efficiency in producing magnetic force
2. Increasing the frequency of the microactuator
3. Increasing the rotation range.
4. Increasing the utilization area.
Efficient Magnetic Microactuator with An Enclosed Core
Efficient Magnetic Microactuator with An Enclosed Core
Improved Design
• 7 layers
- 2 coil layers
- 2 permalloy layers
- 3 insulators
• Dimensions of coils
and plate are dependent
• EM isolation
Design Problem
• Magnetic microactuators with magnetic circuits
- magnetic force varies with given designed dimensions
• This proposed microactuator allows
- a large variation in lengths of poles
The optimal design is to find the optimal values of the geometrical parameters to generate a maximum force
Table 3.1 The ranges of the geometrical parameters (m)
L g p1 p2 c tp tc ti
800 20 0 - 400 0 – 400 0-400 5-50 5-30 40-60
Design Procedure
• First, the effects of geometrical parameters on magnetic
force generation are analyzed by conducting a series of
finite element simulations.
• Then, those geometrical parameters which have critical
effect in optimization are found as design variables.
• Finally, the GA is applied to find the optimal values of the
design variables and the maximum magnetic force.
Model for Magnetic Force Computation
Maxwell equations in the magnetostatic case
The flux density expressed in terms of vector potential
The magnetic co-energy can be calculated as
Using the virtual work principle, magnetic force is
With the equations above, the finite element method is used to solve the problem.
0
1JA
AB
V
dVBdHW
s
WFm
'
Model for Magnetic Force Computation
FEM Model
• 2D axial symmetry element
• 2D infinite boundary element
• unsaturated, r is a constant
• ANSYS software
Initial Design
magnetic force is 512.2 N with the current of 0.08 A
L g p1 p2 c tp tc ti
800 20 200 100 100 10 10 50
Geometrical Parameter Analysis
• six geometrical parameters need to be determined
- pole length p1
- pole length p2
- radius c
- plate thickness tp
- core thickness tc
- insulator thickness h
• with all of six parameters as design variables in optimization, difficult
• analyze the effects of the geometrical parameters - take out those not so critical
- use the remaining critical ones as the design variables
Geometrical Parameter Analysis
Fig. 3.3 Magnetic fluxes at different pole lengths p1:(a) p1=150m, (b) p1=200m, (c) p1=250m, and (d) p1=300m.
Pole Length p1, p2• peak force occurs when pole
length p1 is about 250 m
• Similar results for pole length p2
Fig. 3.2 Relation between pole length p1 and the generated magnetic force.
Geometrical Parameter Analysis
Fig. 3.4 Relation between core radius c and the generated magnetic force.
Magnetic core radius c • peak force obtained when radius
c is about 115 m, the reasons are
in twofold.
- c increases, core reluctance
decreases that helps to increase
the magnetic force.
- c increases, the positions of all
coils move outward which leads
to magnetic force reduction
Geometrical Parameter Analysis
Magnetic core radius c • force generation on different
positions of the single coil
Fig. 3.6 Magnetic fluxes for different coil position(a) inner of the microactuator, (b) under the plate, (c) center of the microactuator, (d) outer of the microactuator.
Fig. 3.5 The influence of the position of a single coil on magnetic force generation.
Geometrical Parameter Analysis
Thickness parameters• larger tp, smaller plate reluctance
• magnetic core thickness tc has the
same phenomenon
• larger the insulator thickness ti,
less the internal leakage flux
The parameters affect magnetic force monotonically
The maximum force is obtained with the maximum thickness
parameters
Pole lengths p1, p2, and radius c are taken as the major
design variables design, to be found by using GA
Developed by Holland, the concept of biological evolution
• multiple search points, not a single point, the probability of
reaching for the global optimum is raised
• do not use any derivative or mathematical information
• nonlinear or unknown systems with a large search space
• Three operators: reproduction, crossover, and mutation
• drawbacks including premature convergence, low search
efficiency, and difficulty for parameter setting
Genetic Algorithm
A. Fitness scaling
to maintain diversity in the population
B. Population-elitist with rank selection reproduction
use the relatively good individuals from the previous
generation
C. Adaptation of operator probabilities
to avoid premature convergence and excessive diversity
Modified Genetic Algorithm
Step 1: Initialize the GA parameters, and generate initial population.
Step 2: Decode each chromosome for design variables and compute each fitness value.Step 3: Execute the fitness scaling.
Step 4: Evaluate each chromosome by performing the population-elitist with rank selection reproduction scheme.
Step 5: Perform the adaptation of the crossover and mutation probabilities.
Step 6: Create the new chromosomes by applying the operations of crossover
and mutation.
Step 7: If not convergent, go to step 2 for the next generation; otherwise, stop and output the optimal values.
Modified Genetic Algorithm
• GA based optimizer that contains a simulator driver to interface with the FEM is developed
• the modified GA includes the three proposed operators,
while the SGA (simple GA) does not
• the modified GA can converge much
more quickly than the SGA
Modified Genetic Algorithm
Fig. 3.9 Comparison of the evolution processes between the SGA and the modified GA.
Effects of GA parameters on the evolution
Modified Genetic Algorithm
number of individuals is better selected as 20
crossover rate is better selected as 60%
mutation rate is better selected as 10%
Results
Fig. 3.11 Magnetic flux distribution for the initial and optimized geometry: (a) initial geometry and (b) optimal geometry.
• The optimal variables are found to be
• Magnetic flux flows much more through
the permalloy plate after optimization
• force is 589.2 N for the optimized model,
larger than 512.2 N for the initial design
the improvement can be achieved by only designing the layout of mask
p1 p2 c
Initial design 200 100 100
Optimal design 290.8 61.1 152.4
Thickness Design
• The maximum force increases as these thickness parameters increase, coincides the previous assumption
• The maximum force approaches the largest value when the plate thickness increases
• core thickness has the most evident effect on maximum force generation
• the relation between the maximum force and insulator thickness is approximately linear
set the thicknesses to be their maximum value simultaneously at 50, 30, and 60 m, maximum magnetic force is 1160.9 N, the largest among all of the models
Results
Macromodel Approach
Generate a Macromodel Directly from 3-D Geometry and Physics
( )( ( )) ( )r
r r
dx tF x t b u t
dt
( ) ( )Tr ry t c x t
Complicated Geometry, Coupled Elastics,
Magnetics
Low order state-space model which captures
input (u)/output(y) behavior
Fig. 4.1 Block diagram of the macromodel approach.
Macromodel Approach
0
ii q
L
q
L
dt
d
),,(),,(),,( tqqUtqqTtqqL
i
mim u
tuuUtuF
),,(
),(*
,
Theoretical Approach
Lagrange’s equations,
L is defined by
T is the kinetic energy and U is the potential energy.
Selecting the meshed nodal displacements u as the generalized coordinate,
and assuming u be the small displacements
M is mass matrix, and Fm is the nodally defined electromagnetic force with
0),( tuFKuuM m
qtqun
iii
1
)(
),( tqFqKqM mTTT
][ , 2i
TT diaqKandIM
),(2 tqFqq mTiii
)()(),( 2 qftItqF mm
Theoretical Approach
Selecting the n-dimensional generalized coordinates,
By introducing the above Eq. into dynamic Eqs. and premultiplying the
result by by T
The basis functions can be determined by using the natural modes,
The dynamic equations become
Fm is proportional to the square of the input current
Theoretical Approach
The equations can be expressed as
is the generalized force, referred to as the force macromodel
On the other hand, the equations derived with the magnetic co-energy [42],
is the magnetic co-energy with unit input current, referred to as the
energy macromodel
The force and energy macromodels are with different computation procedures
)()(22 qptIqq iii
)()( qfqp mTii
i
miii q
qutIqq
)(
)(*
22
)(* qum
Macromodel Generation
Building the approximate closed-form macromodels by
identification technique.
- Sampling a set of the FEM solutions as the fitting data
(experimental design)
- Selecting a model (FLM)
- Fitting the selected model to data (cluster estimation,
backpropagation)
Design of experiments
• n input variables
• The levels are used to adequately span the predetermined input, m levels.
• nm runs or Taguchi’s method
Sampling data
input outputMagnetic Analysisorthogonal array force, energy
Training data L25(56) Testing data L16(4
5)
• In Sugeno-type FLM, the ith rule is described as
• the representation is an integration of the rules rather than a single crisp correlation
• the Gaussian-type membership function
Fuzzy Logic Model
c - c c + Fig. 4.2 Gaussian type membership function.
isoutput rule then the, is and and , is and , is If 2211 innii FxFxFx
niniiini xpxpxppxxxy 2211021 ),,,(
2
2
1exp)(
ij
ijjjijF
cxx
• The weight for each rule’s output becomes
• For the FLM with r rules, the output
can be expressed as
• The differentiation of FLM output
can be analytically derived for
energy macromodel
• The parameters to be determined are
Fuzzy Logic Model
n
j ij
ijjninFiFiFi
cxxxxw
1
2
2211 2
1exp)()()(
r
ii
ii
r
iii
w
wvyvy
1
1
where
ijijiji cpp ,,,0
• Minimize the square of instantaneous error with respect to the unknown parameters
• Gradient-descent method
by applying the chain rule,
• Backpropagation method
Gradient-decent and Backpropagation methods
ijijiji cpp ,,,0
22
2
1)ˆ(
2
1kkkk eyyE kii
i
kii ekvkp
b
Ekpkp )()()()1( 00
0000
jkiijij
kijij xekvkp
b
Ekpkp )()()()1( 11
)(
)()()()()()1(
22*
02 k
kcxykyekvkc
b
Ekckc
ij
ijjkikiij
i
kijij
)(
)()()()()()1(
3
2
30
3 k
kcxykyekvk
b
Ekk
ij
ijjkikiij
i
kijij
2min
2 RR
Magnetic microactuator with complex structure for
demonstrating the efficiency of proposed approach
Geometry:
• an asymmetric 625 x 625 x 5m plate
• four beams 50m wide by 5m thick,
- the shortest 150m, the longest 300-m,
- the others 200m, and 250m
• a 500 x 500 x 5m permalloy
• 32-turn coil, 8 –m thickness for each layer
• 16-m gap
FEM:
• Structure, 411 elements, 799 nodes
• Magnetics, 17408 elements, 19602 nodes
• the resulting deformation is depicted
Simulation Results
• Modal analysis
• QR factorization
• Selecting 6 modes (> 99.5%)
Simulation Results
Mode 1
Mode 8
Mode 6
Mode 10
Mode 2
Mode 3
Mode # Contribution (m)
1 -5.2941
3 0.1789
2 0.0569
6 0.0142
8 0.0044
10 0.0031
7 0.0021
12 0.0021
5 0.0014
9 -0.0013
Mode # Frequency (kHz)
Period (s)
1 16.3 61.46
3 53.5 18.70
2 34.4 29.09
6 192.5 5.20
8 266.3 3.76
10 363.1 2.75
Macromodeling Computation Times
(Pentium III 850MHz)
• The total time would have been longer without experimental design
• FLM identification took only a few minutes, demonstrating its efficiency
Simulation Results
Computation Times (sec) Force macromodel Energy macromodel
Quasi-static analysis 4320 4320
Modal analysis 10 10
Modes selection 30 30
Data sampling 10660 10824
FLM identification 148 16
Total time (hours) 4.21 4.22
Static Simulation
• Force macromodel yielded an error of less than 1.5%
• Energy macromodel shows a much greater error
due to the differentiation of the fitted energy macromodel
Simulation Results
Dynamic Simulation
• Each mode response containing the ripple has the same timing as the applied square waves
• Mode 1 dominated the main response while the other modes reflected the general shape of the applied sawtooth wave
• Each simulation took
about 2 minutes
Simulation Results
0
20
40
appl
ied
curre
nt(m
A)
-5
0
5
mod
e 1
-0.2
0
0.2m
ode
3
-0.1
0
0.1
mod
e 2
-0.01
0
0.01
mod
e 6
-5
0
5x 10
-3
mod
e 8
0 100 200 300 400 500 600 700 800 900 1000-5
0
5x 10
-3
mod
e 10
time (s)
0
20
40
appl
ied
curre
nt(m
A)
-5
0
5
mod
e 1
-0.1
0
0.1
mod
e 3
-0.05
0
0.05
mod
e 2
-0.01
0
0.01
mod
e 6
-2
0
2x 10
-3
mod
e 8
0 100 200 300 400 500 600 700 800 900 1000-2
0
2x 10
-3
mod
e 10
time (s)
Fabrication of Electromagnet
Fig 5.1 A schematic processing sequence for electroplating Cu or NiFe
4. Electroplate Cu or NiFe1. Deposite SiO2
2. deposite seed layer
3. resist layer is spun, exposed, and developed
5.release resist
6. release seed layer
Fabrication of Electromagnet
Fig 5.2 A schematic processing sequence for the fabrication of the coil with enclosed core
6. coate 3rd insulation layer
1. electroplate NiFe
3. electroplate 1st coil
2. coate 1st insulation layer
5. electroplate 2nd coil
4. coate 2nd insulation layer7. electroplate NiFe
Fabrication of Permalloy Plate
Fig 5.3 A schematic processing sequence for the fabrication of the permalloy plate with a 4-suspended-beam structure
1. deposite SiO2 on the double side
2. coate polyimide membrane
3. deposite Au as the mask for the polyimide beams
5. pattern the SiO2 etching window on the backside wafer
4. electroplate NiFe
6. etch in KOH with the Teflon chuck
8. release the polyimide beam with polyimide etch
7. etch SiO2
Results and Discussion
Fig 5.4 Photograph of the electromagnet with the enclosed core: (a) perspective view and (b) top view
• NiFe permalloy is 10m thichness
• Insulator is 10m thichness
6st layer
7st layer, enclosed
• NiFe plate is 800m long and 10m thick.
• The length and width of beam are 1000m and 200m
Results and Discussion
Fig 5.5 Photograph of the NiFe plate supported by 4-suspended-beam structure: (a) perspective view and (b) enlarged view
• Each coil layer includes
- 17 turns
- line width and spacing are 25m
- thickness about 12m.
• The total electrical resistance is approximately 14 Ohm
Results and Discussion
Fig 5.6 Photograph of the Cu coils: (a) the resister pattern and (b) the electroplated coils
• A nanoindentor is used to measure the stiffness of the four-beam structure,
the stiffness k is about 45N/m
• B-H curve as magnetic properties of electroplated NiFe permalloy
Results and Discussion
Fig 5.7 The load-deflection curve of the 4-beam structure Fig 5.8 B-H curve of the NiFe permalloy
• The current-deflection curve is obtained using a laser displacement system
• The test results are well with numerical solution
• 27.6m displacement at 292mA and 4.5V
• The estimated force is bout 1240(N)
Results and Discussion
Fig 5.9 The comparison between the experimental and theoretical results of the current-deflection curves
),(
(2.4) Eq. ),(
(5.1) Eq.
IxFKx
IxFF
KxF
m
mm
s
Conclusions
The proposed magnetic microactuator is ready for practical applications
Crane Vibration Suppression
• Tower cranes are widely used in construction and transportation industries. • The payload acts as a pendulum, operating the crane inappropriately may
result in dangerous situations. • One way to suppress the oscillation is to move the payload slowly, which
leads to inefficiency.
• A control scheme to reduce the oscillation and increase the efficiency of the operation is important.
Tower Crane Model
The position of the payload ,
Lagrange’s equations
The energy T and V are
The velocity
kttLjtLitttLtxtp
(t)cos)(cos)( (t)sin)( ))(sin)(cos)()(()( 21121
21 coscos mgLV )(2
1vvmT
tpkttpv
iLLLLx
121221121 sincoscossinsinsincos
jLxLL
21111 sincoscossin
kLLL 21212211 coscossincoscossin
Tower Crane Model
L
x
L
xL
x
L
x
L
g
L
L
L
L
L
g
L
LL
L
22
1
2
2
2
1
2
1
2
2
2
2
1cos ,sin 111 1cos and ,sin 222
L
x
L
xL
x
L
xx
L
L
L
g
L
L
L
L
L
L
L
gx
22
2 20
0
222
222
1000
0100
TTxxxxx 21214321
Rotary Tower Crane Model
l
sl
s
l
gl
g
22
1
2
2
2
1
2
1
0
02
u
xl
sx
l
gxxxxxux
xl
suxxux
l
gxxx
x
uxf
6
26
3263621
4
6431261
2
2
2
),(
TTxxxxxxx 2211654321
u
Let
),,( uxfx
The state equation becomes
Crane Vibration Suppression
Generalized Input Shaping
Simulation Results
Experiments