Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Models: runner-ups in
system design
Roelof Hamberg, research fellow
Industry-as-Lab
research
projects
Exploratory
research
Consolidation
& transfer
Industrial problem
statements
New competitive
capabilities
Fundamental insights
& expert know-how
Industrial relevant
research drivers
RE
SE
AR
CH
PR
OD
UC
T
DE
VE
LO
PM
EN
TIndustry
Philips, ASML, FEI,
VI, Thales, Océ
Universities
TU/e, UT, TUD,
RU, UU, RUG, UvT
ESI
Industrial
constraints &
scalability
context
2
Outline
3
models&patterns
verify
explore options
generate
But…
this is theory;
let’s start with
three real examples
ASML machine: “Print the Netherlands in 30 seconds”
minimum feature size 30 nmpositional accuracy 5 nm
300 mm
300 mm wafer
300 km
The Netherlands
300 kmx 1 million
corresponds to 3 cmpositional accuracy 0.5 cm
Source: ASML
4
ASML challenge
5
hard real-time (µs) ↓interactions ↑control loops ↑
position
sensors
position
measurement
position
controlactuators
power
amplifiers
position
set-point
wafer stagemotion controlled
predictable timing?
4000 tasks250 sensors/actuators14000 dependencies
Complex electronics platform Many design files & formats
ASML case – approach: performance modeling
platform
model
executable
system model
system
specification
& realization
performance
measurements
performance
prediction
gauging
validation
optimization &
simplification
application
modelless impact
of slack
P1 P2 P3
X X
P1 P2 P3
X X
validation
6
modelingengineering
ASML – performance model results
7
platform
model
executable
system
model
embedded
system
construction
(design)
application
model
mapping
many interacting subsystems4000 tasks250 sensors/actuators14000 dependencies
Manymulti-disciplinary design files & formats
averageworst
latency
ASML benefits
• Product improvement
– 16 different improvement options identified
– over 50% performance gain
– determination of platform limits
• Improvement roadmap
– Cost/risk analysis
– Decision trees
5 µs 4 µs 2 µs 2 µs 7 µs 5 µs 2 µs
3 µs5 µs 1 µs 4 µs 1 µs
1 � ¬1 ☺
11 � 11 �
4 ☺
3 � ¬3 ☺
4 ☺
0
08
11
16.2
12.4
19
3
6.6 3
8.4 6.3
4 ☺
4.8
2 � 9 � 2 �
0
¬2 ☺
short term roadmap
mid term roadmap
8
Océ challenge
9
PDF/PostScript paper
variable timings ↑job dependent scheduling ↑
sufficient throughput?
cost price ↓
sizing?memory?
scheduling?configuration?
building prototype
is expensive: effort
and lead time �
many decisions to
take, behavior
difficult to predict �
don’t think too long, but build!
Analysis
module
Modeling module
(specification)
Kernel module
(intermediate representation)
Diagnostics
module
Search
module
Océ case – tuned towards ease-of-use & exploration
10
trade-offs between system aspects
measurements (logs)compared to
model predictions
Océ benefits
• Early validation beyond spreadsheets
– low investment (< 1 day)
– more insight in dynamics (variability, resource sharing over time, flow control)
– applied in real cases, impact on design decisions
• Multiple properties validated
– 1 specification → multiple models– consistent models
• Design decisions
– guided search through options
11
Vanderlande systems: interwoven in society
12
Imagine:Distribution centre failure
of supermarket at Friday
Consider:New shops, new articles,
Christmas time, …
Vanderlande challenge
13
Goederen Ontvangst en pallet magazijn
OmpakkenDepalletiseren
Opslag lagentrays & ontladen Tray beladen
Rolcontainer beladenHDS picking
Rolcontainer buffer
Verzendgebied
I SS PickingHDS picking
4
2000
Aisl
e 01
Aisl
e 02
Aisl
e 03
Aisl
e 04
Aisl
e 05
Aisl
e 06
Aisl
e 07
Aisl
e 08
42 (27) dc68 (14) sc
a
b b
a
a a
b b
4000 (263 6)
b b b b
Ais
le 01
Ais
le 02
Ais
le 03
Aisl
e 04
Aisl
e 05
Aisl
e 06
Aisl
e 07
Aisl
e 08
Aisl
e 09
Aisl
e 10
Aisl
e 11
3
2
3
2
a
11
11
4000 (12 70)
d d d d
40 40 40 40
40 40 40 40
d d d dd d
d d d d d d d dd d
4x4
4x4
2x 4x4
2x4x4
4x4
4x4
65 35
19
17520
6 6 6
Aisl
e 09
Aisl
e 10
Aisl
e 11
Aisl
e 12
a
2
a
440 (31 8)
4000 (2 945)
240 (159)
3 3
1000 (63 6)
12 00 (1027)
180 0 (1540)
700 (154 )
1
1 20 (107)180 (162)
200 (108)
1 80 (162)Transport naar droog
item picking
3 000 (1010)
(1010 )
500 (440 )
68 (52)
160 (15 7)
150 (1 04)
186 (157)
12 0 (64)
90 (65)
130(65)80 (12)
1 86 (157)
160 (157)
180 (41)160 (6 5)
464
1000 (5 41)
(190)
200 (108) 200 (108)
240 (184) l agen/h
4
300 (8 1)
140 (81) 14 0 (81) 140 (81)
1820
440 (3 18)
3600 (30 81)
3 08
7 00 (616)
4 00
200
838
1141
30435304
796
(32)
2500 250 0
a
4 000 (541)
van droog item picking
140 (81)
79 (53)
600 (28)
188
188
10 m(1 080)
1 0 x
400 (269) ge nest
90 (80)
1c 12
15
11
2 c 14 16
1a 1b
2a 2 b
3b 173c3a 1918
4c20 214a 4b
1a
2a
3a
4a
240 (184 ) lagen /h
4500 (4108)
Aisl
e 12
50 00 (3515)
4 000 (3515)
2 2
d d
Aisl
e 03
Aisl
e 04
Aisl
e 05
Aisle
06
Aisle
07
Aisle
08
Aisle
09
Aisle
10
Aisle
11
Aisle
12
Aisl
e 13
Aisle
14
Aisle
15
Aisl
e 16
Aisl
e 17
Aisl
e 18
Aisle
19
Aisle
20
Aisl
e 21
Aisle
22
Aisle
23
Aisl
e 24
1113 15 17 19 21
1 500 (1027 )
1500 (1027)
1500 (1027)
12 14 16 18 20 22
Aisl
e 01
Aisl
e 02
Aisl
e 03
Aisle
04
Aisle
05
Aisle
06
Aisle
07
Aisle
08
Aisle
09
Aisle
10
Aisle
11
Aisle
12
Aisl
e 13
Aisle
14
Aisle
15
Aisl
e 16
Aisl
e 17
Aisl
e 18
Aisle
19
Aisle
20
Aisl
e 21
Aisle
22
Aisle
23
Aisl
e 24
1b
2b
3b
4b
1c
2c
3c
4c
13
2 40 (158) l agen/ h
240 (1 59)
90
300 0 (978)
90
90
4000 (3.08 1)
C a. 10. 000
450 0 (4108)
1500
Aisl
e 01
Aisl
e 02
1 500 (1027)
3
3
8900
Aisl
e 01
Aisl
e 02
3
3
1 500 (292)
Miniload “st aart”
1500
3 shuttles
3 shuttles
300 (257)
40 00 (3515)
customer data dependencies ↑unscheduled down time ↓variability in business processes ↑
sufficient throughput?
availability?first time right laborious layout
can you make a
good design in
short time?
Vanderlande – model approach & results
14
scripts
statistics foro throughputo work-in-processo lead times
visualization of schematic system
Vanderlande benefits
• Early validation of design parameters
– system sizing, buffer sizes, control rules
– visualization to verify behavior
– full layout not needed
• Customer system configuration
– towards toolbox supportcustomer data → filter → concept simulation
– standardized processes and design steps
15
so...
what have we seen?
some reflections
16
Early validation of system aspects
17
verify
• visualization of candidate systems
• non-technical stakeholders
• increased insight →recipe for growth
Excel++: Executable system models
18
• overcomestatic limitations
• average → optimistic
• worst → pessimistic
• spreadsheet
real product
• documentation
dynamicmodels
Think before you act
19
chose from many options
• evidence-baseddesign choicesbefore prototyping
• increased insightin parameter effects
• model structure
• smart search &optimization
Design process efficiency
20
generation
• generate system from modelthe model is your system
• need for stable architecture
• standardized process
Models: runner-ups in
system design
21
Embedded Systems InstituteEindhovenwww.esi.nl