optimality of monotone procedure
TRANSCRIPT
The University of Electro-CommunicaionsFaculty of Electro-Communications
Department of Systems EngineeringSuzuki Lab
Dinesh Rajapaksha2013-02-19
1
Optimality of Monotone Procedure for a System Working Under
Variable Operating Conditions
Background
Systems deteriorate over time when used
Companies want the maximum performance from their systems during its life time
Optimal decisions made in a short period of time in operation will give the maximum performance
2
http://www.intel.com/pressroom/archive/releases/2006/20060125comp.htm(2013/02/15 access)
Plasma Enhanced-CVD[1]
3
WaferChemical Layers
Semiconductor Industry
Diagram of Plasma Enhanced-CVD (Chemical Vapor depositor)[2]
Plasma
Gases
G
a
S
4
Wafer
Chemical Layers deposited
Plasma
Gases
Wafer
“System” and “Operating Condition”
5
SYSTEM
Wafer type 1
Wafer type 2
Wafer type 3
Operations needed ioProduct
)1(.ConditionOp
)2(.ConditionOp
)3(.ConditionOp
Operating Conditions
Plasma Enhanced-CVD
Which Operating Condition to use ? When to Use ?
To get the maximum cost performance ofPlasma Enhanced-CVD during it’s life time
…0 1 i2 nNew
System’s state
Problem in decision making
10nStates of the system :
Available operating conditions : 3h
System
There are possible decisions available.59049310nh
Method: Monotone procedure 6610121310 CCH nnh
6
Decision Combinations (repeated combinations)
Purpose of the research
Purpose of this research is to providesufficient conditions for optimal decisionmaking to be given by a monotoneprocedure.
7
Previous research
8
C. Deman, et.al[3] M. Ohnishi, et.al[4] This research
System’s state Directly observable Partially observable Directly observable
Options toDeicide
Replace, Keep Replace , Inspection, Keep
h Different operatingconditions
…0 1 i2 nNew FailureSystem’s state
ReplaceKeep
…0 1 i2 nNew FailureSystem’s state
Keep ReplaceInspection Keep
Denotation
9
: State space of system;
: True state of system;
: Operating condition space of system;
: Operating condition;
: Transition probability of system state from i to j when
: Transition probability matrix of system;
: Operation cost per 1-period at when and
: Discount factor 0 < < 1
},...,1,0{ nS
SX
},...,1{ h
O
kO
Sji
k
ij
k p ,}{P
iS
kO
S
X
O
k
ijp
kP
k
iC
Product 1
SYSTEM
Op. Condition 1 Op. Condition 2 Op. Condition 3
10When do we select the suitable Operating condition ?
Model Description
deterioration
Product 2Product 3
)2()2()2(
)2()2()2(
)2()2()2(
0
0
000
nnn
iniji
j
ppp
ppp
ppp
nj
n
curren
t state
next transition state
1P
2P
3P
)()()(
)()()(
)()()(
0
0
000
hh
nj
h
hhh
h
n
hh
nnn
iniji
j
ppp
ppp
ppp
)1()1()1(
)1()1()1(
)1()1()1(
0
0
000
nnn
iniji
j
ppp
ppp
ppp
nj
n
),()(
)3,()(
)2,()(
)1,()(
min)(
0
)()(
0
)3()3(
0
)2()2(
0
)1()1(
hivjVpc
ivjVpc
ivjVpc
ivjVpc
iV
n
j
h
ij
h
i
n
j
iji
n
j
iji
n
j
iji
when using Operating condition 1
),()(min)(0
)()( kivjVpciVn
j
k
ij
k
i hk ,...,2,1
Model Formulation
discount factor 10 11
when using Operating condition 2
when using Operating condition 3
Total discounted cost function
when using Operating condition h
Assumptions
A-1
A-2
A-3
A-4
A-5
)()1()( icc k
i
k
i
hkSIP k ,......2,1)(
)()( ic k
i
SIPP kk )1()(
12
)(
0
)(
0
)1(
0 ...... hk ccc
Properties of the Optimal Cost function
)0(),0(....)2,0()1,0( ihvvv
)(),( ikiv
)()1,(),( ikivkiv
1)
2)
3)
v(0,1)
v(0,2)
v(0,3)
State i
Cost
Is a non-decreasing function of i
),( kiv v(i,k)
v(i,k+1)
13
0
14
deterioration
Cost
Op Cond3Op Cond1
Op Cond2
Op Cond3
L0New
Op Cond3
Op Cond2
Cost function for Op condition1
Cost function for Op Conditon3
Cost function for Op condition2
A System with 3 Variable Operating Conditions
Monotone Procedure
Op Cond1 Op Cond2 Op Cond3
Example
conclusion
• Systems that follow Assumption (A-1) to (A-5),
the optimal decision procedure can be given by a monotone procedure.
Op Cond2 Op Cond2 Op Cond3
State i
cost
Op Cond1
Op Cond2
Op Cond3
• Reference
[1] 谷口研二(2004)「LSI設計者のためのCMOSアナログ回路入門」 pp.22-29.
[2]https://www.crystec.com/trioxide.htm (2013/01/20 access)
[3] C.Derman(1963): “On optimal replacement rules when changes of state
are Markovian”, Mathematical optimization techniques,University of California Press.
[4] M.Ohnishi, H.Kawai and H.Mine, (1986): “An optimal inspection and replacement policy under incomplete state information” European Journal of Operations Research , vol.27, pp.117-128.
16
Thank You
)()()(
)()()(
)()()(
0
0
000
aa
nj
a
aaa
a
n
aa
a
nnn
iniji
j
ppp
ppp
ppp
P
18
A-3A-4A-5
i
k
iC
iC k
i
)(
)(),,1,0(1 iinkppSISIn
kj
s
ji
n
kj
s
ij
skk PPP
A-2
)( SI EaaP
≧≧
A-1
increasing Stochastic:SI
k
iC
1k
iC
iCC k
i
k
i
)1()(
)(
0
)1(
0
)0(
0 .. hCCC
Suppose that we have a set of Operating Condition Options
Op. Cond(1) < Op. Cond(2) < ……. < Option(h)
Option(x)Option(x)
Option(x)Option(x)
Option(x)Option(x)
operating conditions set
)1(.)(. kConditionOpkConditionOp ),...2,1( Nk
cost
Op. Condition number k
19
W1W4
W4W3
W3W2
W2
Ordering Operating Conditions
Temp.
Speed
Chemicals
Op. Condition(x)
Example
W1
SETTING
Mass Production
Ordering the productions with MULTIPLE Plasma Enhanced-CVDs
CVD
W1W1
W1W1
W1W1
W1
CVD
W1W1
W1W1
W1W1
W2
W3
CVD
W3W3
W3W3
W3W3 Op condition1
Op condition2 Op condition 3
State i
cost
Op condition 1
Op condition 2
Op condition 3
W3
CVD
W3W3
W3W3
W3W3
20
Cost function of system working underOperating condition k
SYSTEM
Op. Condition 1
Productions
n
j
k
ij
k
i jVpckiv0
)()( )(),(
),( kiv
State i
Cost
),( kiv
0
Cost function of the system when operating condition k and state is i
21
Numerical Example
When Parameters follow assumptions made
7.02.01.0
2.03.05.0
1.02.07.0
4.03.03.0
2.02.06.0
1.01.08.0
2.03.05.0
1.02.07.0
05.005.09.0
9.0
2N
k
ic101
40
20
1
2
1
1
1
0
c
c
c
103
42
23
2
2
2
1
2
0
c
c
c
105
45
28
3
2
3
1
3
0
c
c
c
SIP )1( SIP )2( SIP )3(
Discount factor
number of periods considered
Transition probability matrix for each Option
One period operating cost for each Option
22
result
48.8
57.8
92
52.43
61.16
76.91
57.43
63.19
74.78
0
10
20
30
40
50
60
70
80
90
100
Op. Condition1 Op. Condition2 Op. Condition 3
Cost
State
Numerical Example
When Parameters NOT follow assumption A-4
7.02.01.0
2.03.05.0
1.02.07.0
4.03.03.0
2.02.06.0
1.01.08.0
2.03.05.0
1.02.07.0
05.005.09.0
9.0
2N
k
ic100
40
30
1
2
1
1
1
0
c
c
c
95
35
25
2
2
2
1
2
0
c
c
c
85
25
15
3
2
3
1
3
0
c
c
c
SIP )1( SIP )2( SIP )3(
Discount factor
Number of periods considered
Transition probability matrix for each Option
One period operating cost for each Option
24
Cost
State
result
65.1
82.3
172.9
54.7
71.9
135.4
32.1
46.6
113.9
0
20
40
60
80
100
120
140
160
180
200
Op. Condition1 Op. Condition2 Op. Condition3