[系列活動] 製造資料科學:從預測性思維到處方性決策
-
date post
18-Feb-2017 -
Category
Data & Analytics
-
view
2.837 -
download
5
Transcript of [系列活動] 製造資料科學:從預測性思維到處方性決策
-
Data Science in Manufacturing
From Predictive to Prescriptive
Dr. Chia-Yen Lee ( )
Institute of Manufacturing Information and Systems () Dept. of Computer Science and Information Engineering ()
Engineering Management Graduate Program () National Cheng Kung University ()
-
@NCKU ()
Outline
2
1
Intelligent Manufacturing Systems
2
Characteristics of Manufacturing Dataset and Data Preprocessing
3
Cases, Issues and Myths
4
From Predictive to Prescriptive
-
@NCKU ()
Intelligent Manufacturing Systems
3
-
@NCKU () 4
Overview
Intelligent
Manufacturing
Systems
What is Systems?
What is Manufacturing?
What is Intelligent?
-
@NCKU () 5
What is Systems?
Systems is
Process Input Output
Feedback & Control
Environment
Boundary
-
@NCKU () 6
Human Body System
Human Body
Inputs (Five Sense)
Hearing
Sight
Touch
Smell
Taste
Process
Learning
Boundary
Environment
Good Outputs
Mental Growth
Physical Growth
Bad Outputs:
Pressure
Urine, Excrement
Exhale: CO2
-
@NCKU () 7
What is Manufacturing?
Manufacturing is Manufacturing is
Manufacturing is the realization () of product. Swann (2003)
-
@NCKU ()
How to throw a paper farther?
8
-
@NCKU ()
Manufacturing is
9
Idea? (by brain-storming and association)
-
@NCKU ()
Manufacturing is
Manufacturing Process (Handmade)
For your needs (function)
10
(2011).. . http://www.zwbk.org/MyLemmaShow.aspx?zh=zh-tw&lid=150882
-
@NCKU ()
obj1. How to throw a paper farther?
obj2. The product should be delightful.
(this is what we call the customer requirements)
11
?
-
@NCKU ()
Valued added?
Material change
12
google
-
@NCKU () 13
Valued added?
Process change
google
-
@NCKU () 14
Valued added?
Process change
google
-
@NCKU () 15
Valued added?
Process change
google
-
@NCKU ()
?
16
Customer Requirements
Obj1. How to throw a paper farther?
Obj2. The product should be delightful.
Obj3. Low cost ()
Obj4. Specification ()
Obj5. Firmness ()
.
...
..
KPI
-
@NCKU ()
Performance Index
17
Yurdakul, Mustafa (2002),"Measuring a manufacturing systems performance using Saatys system with feedback
approach", Integrated Manufacturing Systems, Vol. 13 Iss 1 pp. 25 - 34
-
@NCKU ()
What is Intelligent?
18
-
@NCKU () 19
-
@NCKU ()
Intelligent
20
Shutterstock (www.shutterstock.com)
-
@NCKU ()
Knowledge? Experience?
Any others?
21
-
@NCKU () 22
Knowledge
-
@NCKU ()
: ? ?
email
23
Advanced Analytics Intel: SETFI: Manufacturing data: Semiconductor tool fault isolation. Causality Workbench Repository,
http://www.causality.inf.ethz.ch/ repository.php (2008)
-
@NCKU ()
: ? ?
24
! !
! !
! !
!
-
@NCKU ()
Transformation from Data, Information, to Knowledge
25
Etu (2013)
-
@NCKU () 26
Experience
Experience Decision Problem
-
@NCKU () 27
()
http://www.herogamingjobs.com/2014/01/07/experience-vs-knowledge/
Wisdom is
-
@NCKU () 28
Combine together is
Intelligent + Manufacturing + Systems = ?
-
@NCKU () 29
Definitions
IMS is a system which improves productivity by systematizing the
intellectual aspect involved in manufacturing and flexibility by integrating
the entire range of corporate activities so as to foster the optimum in
relationship between men and intelligent machines (H. Yoshikawa,
University of Tokyo).
An intelligent manufacturing process has the ability to self-regulate
and/or self-control to manufacture the product within its design
specifications (Nam Suh, MIT)
Definition in this lecture
is a decision-oriented system which has the computational
intelligence and self-learning ability to optimize the
manufacturing process.
Definition of IMS
-
@NCKU () 30
IMS Framework
Tolga Bozdana (2012)
-
@NCKU ()
Manufacturing Intelligence
()
31
-
@NCKU () 32
Purpose
The purpose of business is to create and keep a customer.
(Peter Drucker)
The social responsibility of business is to increase its profits.
(Milton Friedman)
Method
Transform the resource to product (or service) and by-product
Nature
a portfolio of resources (land, building, labor, equipment, electricity, etc.)
deployed through a network of activities
to create value for a customer
in order to create economic profit () for its owners
The Manufacturing Enterprise
-
@NCKU ()
Value and Profit
Value
Manufacturing is value-added process for the realization of the product
Value Price Cost (VPC) framework
Customer Surplus vs. Producer Surplus
33
Price
Value
Cost
Consumer
Surplus
Producer
Surplus
Wikipedia (2016)
-
@NCKU () 34
Business Decision Framework
Pricing
Strategy
Demand
Planning
Capacity
Portfolio
Cost
Structure
Price Elasticity
Product Mix
Utilization Demand
Fulfillment
Cost Parity
Capital Expenditure
Structure
Profitability Margin Return
ROIC, Payback
(Chien and Zheng, 2011)
-
@NCKU ()
e-Manufacturing Hierarchy
35
International SEMATECH (2002)
-
@NCKU ()
36
Source: International SEMATECH e-Diagnostics and EEC Guidance 2003
Equipment Health Monitoring,
Predictive Maintenance (PdM)
Statistical Process Control,
Virtual Metrology, RMS
Carbon
Emission
RFID, Vision Guidance
Energy
Mgmt
Proactive Decision
(Forecasting)
-
@NCKU ()
Manufacturing Networks
37
Stage1 Stage2 Stage3 Stage4 Stage5
Material
Warehouse
Stocker
(Buffer) Shipments
(Inspection)
work-in-process
(WIP)
Picking
Feeding
Pack&Ship
Bottleneck ()
-
@NCKU ()
Layout and AMHS
38
Interbay Transport
Interbay Transport
ITS
ITS
ITS
ITS
ITS
ITS
Inter-floor
Transfer Stocker
EQ1
EQ1
EQ1
EQ2
EQ1
EQ1
EQ1
EQ2
EQ4
EQ4
EQ3
EQ2
EQ4
EQ4
EQ3
EQ2
EQ5
EQ5
EQ5
EQ2
EQ5
EQ5
EQ5
EQ2
Lee et al. (2014)
-
@NCKU () 39
Strategic Level
Financial PlanningStrategic Business
Planning
Marketing
Planning
Make-or-Buy
Strategy
Tactical Level
Aggregate Production
Planning
Vendor Selection &
Order Allocation
Master Production
Schedule
Resource Planning
Rough-Cut Capacity
Planning
Capacity Require.
Planning
Material Require.
Planning
Monthly Material
Planning
Capacity Planning
Productivity &
Efficiency Analysis
Vendor Relationship
Mgmt.Demand Mgmt.
Outsourcing Planning
Purchasing
Vendor Scheduling &
Daily Assignment
Order Releasing
Shop Floor Operation
Scheduling
Shop Floor Control &
Data CollectionFollow-Up
Production
Activity ControlIn-house Outsourcing
Operational Level
Logistics Mgmt.
Long Term
Mid Term
Short Term
Facility Layout
Production Management
Lee and Johnson (2013)
-
@NCKU () 40
Rapidly changing uncertain environment
Demand and market conditions
Diversity of customer requirements in global markets
Demands for mass customization/product variety
Short product life cycles
Need flexibility and shorter time-to-market
Short Innovation/Product Development Cycles
Inventory becomes a major risk
Push the manufacturing industry service industry
Cost of Capacity
A new 12 semiconductor fab may cost US$3-5billion
Idle capacity is a CEOs nightmare
Why Need Production Management?
-
@NCKU () 41
Manufacturing system vs. Service system Manufacturing: sales quantities and prices are defined before production
due to a longer production lead time (Internal Demand)
Service: non-storable commodities which once transformed from inputs, must be consumed by customers immediately (External Demand)
Main difference is Inventory!
Inventory
Raw materials, Components, Work-in-process (WIP), Finished goods
Lead Time + Uncertainty = Inventory
Inventory Reduction
Reduce material and production lead time (includes transport)
Reduce information delay times (, Bullwhip effect)
Improve quality of information (reduce uncertainty)
Manufacturing system vs. Service system
-
@NCKU () 42
Difference between Manufacturing and Service
Now, manufacturing system moves toward service system.
Manufacturing vs. Service
System Manufacturing Service
Input Material Labor
Process Capital-intensive Build SOP
Labor-intensive Building SOP is difficult
Output Physical product
Intangible Non-separable Non-stable Non-storable Time-concerned
feedback Quality Control Performance criteria
Difficult to measure quality and performance
-
@NCKU () 43
Chien et al. (2004)
-
@NCKU ()
Why is it difficult to manage a
manufacturing systems?
44
-
@NCKU ()
Diverse Resources and KPIs (KPI)
45
(2011)
-
@NCKU ()
(ICT )
46
Prof.Dr. Engelbert Westkmper, Fraunhofer IPA Stuttgart, Germany, Factories of the Future beyond 2013: The role of ICT
http://cordis.europa.eu/fp7/ict/micro-nanosystems/docs/fof-beyond-2013-workshop/westkaemper-manufuture_en.pdf
!? ()
()
-
@NCKU () 47
Variability is anything that causes the system to depart from
regular, predictable behavior.
Sources of Variability:
setups workpace variation
machine failures differential skill levels
materials shortages engineering change orders
yield loss customer orders
rework product differentiation
operator unavailability material handling
Variability ()
Variability from Resource!
-
@NCKU ()
: ??
(variability)?
48
-
@NCKU ()
Transparency- showing the uncertainties in MFG
49
Lee, Jay and Lapira, Edzel (2013). Recent Advances and Trends in Predictive Manufacturing in Industry 4.0
Environment. http://reliabilityweb.com/articles/entry/recent_advances_and_trends_in_predictive_manufacturing/
Internal Uncertainties
Visible
machine failure
product defects
long time delays
drops in overall equipment
effectiveness (OEE)
standardization
Invisible
machine degradation
component wear, etc.
predictive and monitoring
External Uncertainties
-
@NCKU ()
MES
APC
FDC
SPC
VM
R2R
RMS
Preventive Maintenance
Predictive Maintenance
Yield Management
etc
Demand Forecast
Long-Run Capacity Planning
Short-Run Capacity Planning
CapEx
Cost Structure
MPS & MRP
Scheduling & Dispatching
Inventory Management
Order Releasing
etc
Quality
Productivity
Cost Down
Cycle Time
50
-
@NCKU ()
CIM and Automation
51
(Wang, 2012)
-
@NCKU ()
52
-
@NCKU ()
53
-
@NCKU ()
Lean (identify non-value-added process and remove it)
Value
Value Stream Mapping (VSM)
Waste elimination (7 muda, Womack and Jones, 2003)
1. Transportation
2. Inventory
3. Motion
4. Waiting
5. Overproduction
6. Overprocessing
7. Defects
Continuous flow
Line Balancing
Pull production system
Lean Manufacturing
54
-
@NCKU ()
PDCA cycle
PDCA cycle (Deming, 1950s)
PLAN: Establish the objectives and processes necessary to deliver
results in accordance with the expected output (the target or goals).
DO: Implement the plan, execute the process, make the product.
CHECK: Study the actual results (measured and collected in "DO"
above) and compare against the expected results (targets or goals
from the "PLAN") to ascertain any differences.
ACT: Request corrective actions on significant differences between
actual and planned results.
55
-
@NCKU ()
Continuous Improvement by PDCA
56
(standardization) (institutionalization)
cost down
Wikiwand (2016). http://www.wikiwand.com/en/PDCA
-
@NCKU ()
Cost Down?
Cost Down!?
57
-
@NCKU () 58
Intelligent Factory is a decision-oriented system which has the
computational intelligence and self-learning ability to optimize the
manufacturing process.
Based on Data ()
Real-time Feedback Control ()
-
@NCKU () 59
generated by Wordle (2012)
Really!?
-
KNOWLEDGE DISCOVERY IN DATABASES (KDD)
Data
Warehouse
Knowledge
Source :From Data Mining to Knowledge Discovery: An Overview, Advances in Knowledge Discovery and Data Mining, AAAI Press/The MIT Press 1996.
Selection
Preprocessing
Target Data
Preprocessed
Data
Pattern
Transformed
Data
Data Mining
Transformation
Interpretation/
Evaluation
ETL (Extract-Transform-Load)
Storage and Calculation
IT Infrastructure
Pattern Extraction
Value Interpretation for Profitability
Business Applications
60
-
@NCKU ()
61
-
@NCKU () 62
In God we trust, all others must bring data - Edward Deming (1900-1993)
What gets measured, gets managed - Peter Drucker (1909-2005)
-
@NCKU ()
A Short Break
Q&A
63
-
@NCKU ()
Characteristics of Manufacturing
Dataset and Data Preprocessing
64
-
@NCKU ()
Manufacturing Evolution
65
3.04.0 1. 2.() 3.()
-
@NCKU () 66
How to improve a complex
manufacturing system?
Pittsburgh Technology Council (2014). http://www.pghtech.org/media/64942/panoupdated.jpg
-
@NCKU ()
Get to the Bottom of Our
Problems
()
(The Characteristics of Manufacturing)
67
-
@NCKU ()
Characteristics of Manufacturing
The Nature of Manufacturing Systems
Multiple Resources Management
Cause: diverse product
Effect: complicated management (eg. inventory, human resource, etc.)
Network Systems
Cause: different and multiple processes
Effect: interrelationship between upstream and downstream
(eg. process control, WIP, setup)
Dynamic Adjustment
Cause: demand fluctuation product-mix change
Effect: bottleneck shift, WIP
Dynamic bottleneck shift
68
-
@NCKU ()
The Nature of Manufacturing Systems
Production Lead-Time (cycle time, CT)
Cause: material procurement and processing time
Effect: time-to-market
Throughput (TH) of Production Line
Cause: Different process or new-and-old machines (i.e., vintage issue)
Effect: bottleneck and machine capability ()
Littles Law: WIP = TH x CT
Cost-based system (for the same product)
Cause: limited resource to satisfy the customer requirement
Effect: driving productivity and cost down
Note: process or design improvement value-based (for new product)
Note: pure MFG doesnt involve the pricing decision
69
-
@NCKU ()
Manufacturing Field (problem clarification)
Ask Why in shop floor level ()?
()!?
No support (labor/machine/material/management/supplier/inventory)
Manufacturing Dataset (data for analysisbut no data?)
Ask Why?
I dont knowno idea
Long long time ago
No one take over the
None of my business
Please ask somebody/window/dept./supplier/business(competitor!?)
I am the newcomer
Data is not accurate because
70
-
@NCKU ()
71
Characteristics Issues
Batch size () Lot ID decomposition, trace back, yield calculation
Parallel machine () Missing value
Golden machine () Utilization, class imbalance Inference bias
Recipe and parts ()
Nominal or categorical variable too many class too many dummy variables
Sampling testing () Missing value
Engineering or R&D lot ()
Outlier, machine contamination, setup capacity loss
Maintenance ()
When? How ( or )? How much? Capacity loss, Reliability
Changeover () Setup time, capacity loss
Bottleneck shift() Different treatment, WIP transfer
Queue time limit() Defects, WIP
Data imbalance () Inference bias Inventory = Lead Time + Uncertainty
-
@NCKU () 72
() Data Preparation
Data Merge (trace back for recipe diagnosis)
Lot ID Sub Lot ID Final inspection Pass or Fail
Lot001 Lot001 Pass
Lot002 Lot002_1 Pass
Lot002 Lot002_2 Fail
Lot003 Lot003_1_1 Pass
Lot003 Lot003_1_2 Pass
Lot003 Lot003_2 Pass
Lot ID WS1_A WS1_B WS2_A WS2_B
Lot002
Sub Lot ID WS7_A WS7_B
Lot002_1
Lot002_2
R&D
-
@NCKU ()
() (yield calculation)
73
Lot ID Pass or Fail
Lot001 Pass
Lot002 Pass
Lot003 Fail
Lot004 Pass
Lot ID Pass or Fail Lot size
Lot001 Pass 20
Lot002 Pass 25
Lot003 Fail 10
Lot004 Pass 15
75%
85.7%
R&D
-
@NCKU ()
Parallel Machine
Not identical () Tool Matching
Data Preparation
Missing Value
74
Lot ID WS_A_ Mach_1_Temp
WS_A_ Mach_2_Temp
Lot001 820 N/A
Lot002 820 N/A
Lot003 N/A 840
Lot004 N/A 840
Lot ID WS_A_ Temp
WS_A_ Mach_Type
Lot001 820 1
Lot002 820 1
Lot003 840 2
Lot004 840 2
WS_A_Mach_1
WS_A_Mach_2
R&D
-
@NCKU ()
Golden Machine
(inference bias)
75
Lot ID WS_A_ Mach_1_Temp
WS_A_ Mach_2_Temp
Lot001 820 N/A
Lot002 820 N/A
Lot003 820 N/A
Lot004 830 N/A
Lot005 820 N/A
Lot006 820 N/A
Lot007 825 N/A
Lot008 830 120
Lot009 820 130
Lot010 N/A 120
Lot ID WS_A_ Mach_1_Temp
WS_A_ Mach_2_Temp
Mach_1 Mach_2
Lot001 820 Avg. 1 0
Lot002 820 Avg. 1 0
Lot003 820 Avg. 1 0
Lot004 830 Avg. 1 0
Lot005 820 Avg. 1 0
Lot006 820 Av.g. 1 0
Lot007 825 Avg. 1 0
Lot008 830 120 1 1
Lot009 820 130 1 1
Lot010 Avg. 120 0 1
R&D
-
@NCKU ()
Recipe/ Parts- Nominal () or Categorical () Variable Transfer to dummy variable (, )
Given N levels, the method will generate N-1 dummy variables.
76
Lot ID WS_A_ Mach_1_Parts
Lot001 PartsA
Lot002 PartsB
Lot003 PartsA
Lot004 PartsC
Lot005 PartsD
Lot006 PartsE
Lot007 PartsE
Lot008 PartsA
Lot009 PartsC
Lot010 PartsE
Lot ID WS_A_ Mach_1_ PartsA
WS_A_ Mach_1_ PartsB
WS_A_ Mach_1_ PartsC
WS_A_ Mach_1_ PartsD
Lot001 1 0 0 0
Lot002 0 1 0 0
Lot003 1 0 0 0
Lot004 0 0 1 0
Lot005 0 0 0 1
Lot006 0 0 0 0
Lot007 0 0 0 0
Lot008 1 0 0 0
Lot009 0 0 1 0
Lot010 0 0 0 0
R&D
-
@NCKU ()
level (Recipe or Parts) Dummy Variables
Issue: Curse of Dimensionality
77
R&D
-
@NCKU ()
78
R&D
()
NO.001 A 13.723 7.235
NO.002 B NA NA
NO.003 B 13.728 7.237
.
.
.
NO.099 B NA NA
NO.100 A 13.726 7.236
X Y ()
Skills of missing value imputation:
1. // 2. K-Nearest Neighbours
3. Prediction Model http://www.theanalysisfactor.com/seven-ways-to-make-up-data-
common-methods-to-imputing-missing-data/
-
@NCKU ()
Outlier
LotIDLotIDrecipe
79
()
R&D
-
@NCKU ()
(10,000)
updown (Overall Equipment Effectiveness, OEE)
queue time
(eg. status variable identification, SVID)
()
80
STA
GE
x_S
VID
x
0 5
10
1
5
2
0
2
5
0 200 400 600 800 1000 Lot or
Time
R&D
-
@NCKU ()
/
/1product
81
ID Product
Type
Process
Time (hrs)
Start
Time
Lot1 A 1 2016/12/26 14:50
Lot2 A 1 2016/12/26 15:50
Lot3 B 2 2016/12/26 17:50
Lot4 B 2 2016/12/26 19:50
Lot5 B 2 2016/12/26 21:50
Lot6 C 1 2016/12/27 00:50
Lot7 C 1 2016/12/27 01:50
R&D
-
@NCKU ()
(bottleneck)
WIP
//
//
/
82
R&D
120
units/day
140
units/day
80
units/day
120
units/day
100
units/day
80 units/day
()
(2016)
-
@NCKU ()
(Queue Time Limit)
(defect)
Queue Time LimitFOUP
(, furnace)(Form Batch)(recipe)
Qtime(x)(y)Qtime
estimated by the difference between check-out of A and check-in of B
83
R&D
-
@NCKU ()
(Queue Time Limit) Astage1
84
ID Product
Type Stage
Process
Time (hrs)
Start
Time
Lot1 A 1 1 2016/12/26 14:50
Lot1 A 2 1 2016/12/26 16:00
Lot1 A 3 0.5 2016/12/26 17:20
Lot1 A 4 1 2016/12/26 18:00
Lot1 A 5 0.5 2016/12/26 20:30
1 Alarm!!
R&D
-
@NCKU ()
, i.e., KEY
KeyLot ID
?
85
Time SVID 101 SVID102 ..
2/11 00:00:00
2/11 01:00:00
2/11 02:00:00
.
.
.
2/11 23:00:00
Time SVID 1 SVID 2 ..
2/11 00:06:29
2/11 00:10:41
2/11 03:41:09
.
.
.
2/11 23:11:57
Periodic-based record Event-based record
R&D
-
@NCKU ()
Data Merge
86
R&D
Dong and Lee (2017)
Event Periodic
1
()
(Event)
Nearest time
Rolling Forward
Rolling Backward
Nearest time
-
@NCKU ()
Date A
1 2016-01-01 A1
2 2016-04-01 A2
3 2016-07-01 A3
4 2016-09-15 A4
Date B
1 2016-02-20 B1
2 2016-05-01 B2
3 2016-06-15 B3
4 2016-07-01 B4
5 2016-12-31 B5
Date A B
1 2016-01-01 A1
2 2016-04-01 A2 B1
3 2016-07-01 A3 B4
4 2016-09-15 A4 B4
A1 A2 A3 A4
B1 B2 B3 B4 B5
A
B
B
B1 B4 B4
- Rolling forward
R&D
87
-
@NCKU ()
Date A B
1 2016-01-01 A1 B1
2 2016-04-01 A2 B2
3 2016-07-01 A3 B4
4 2016-09-15 A4 B5
A1 A2 A3 A4
B1 B2 B3 B4 B5
A
B
B
B1 B2 B4 B5
- Rolling backward
R&D
Date A
1 2016-01-01 A1
2 2016-04-01 A2
3 2016-07-01 A3
4 2016-09-15 A4
Date B
1 2016-02-20 B1
2 2016-05-01 B2
3 2016-06-15 B3
4 2016-07-01 B4
5 2016-12-31 B5
88
-
@NCKU ()
Data/Class Imbalance # of qualified product extremely dominates the # of defective product
() /
? For the two classes (0 and 1), rule of thumb
10% vs. 90%? 5% vs 95%? or 1% vs. 99%?
It depends on your industry applications.
From a theoretical viewpoint, it occurs if it skews the model training for
prediction
Overfitting? Class Imbalance?
89
R&D
-
@NCKU () 90
Lot ID X1 X100 Inspection
Lot01 PASS
Lot02 PASS
Lot03 PASS
Lot04 PASS
Lot05 PASS
Lot06 PASS
Lot07 FAIL
Lot08 PASS
Lot09 PASS
Lot10 PASS
Lot11 PASS
Lot12 PASS
Inspection
1FAIL
PASS
X1~X100
11/12 = 91.7%
R&D
-
@NCKU ()
91
-
@NCKU ()
200100?
92
-
@NCKU ()
(Scale)
93
(nominal scale)
(categorical
scale)
(ordinal scale)
Bin
(interval scale)
(ratio scale)
(absolute scale)
(2014)
-
@NCKU ()
94
()
(2014)
-
@NCKU ()
: ?
?
95
-
@NCKU ()
: ?
96
11
1 0.0886 0.0886 0.0886 0.0886
2 0.0684 0.0684 0.0684 0.0684
3 0.3515 0.3515 0.3515 0.3515
4 0.9874 0.9874 0.9874 0.9874
5 0.4713 0.4713 0.4713 0.4713
6 0.6115 0.6115 0.6115 0.6115
7 0.2573 0.2573 0.2573 0.2573
8 0.2914 0.2914 0.2914 0.2914
9 0.1662 0.1662 0.1662 0.1662
10 0.44 0.44 0.44 0.44
11 0.6939 ? 0.3731 0.6622
0.4023 0.3731 0.3731 0.3994
0.2785 0.2753 0.2612 0.2753
0.3208 0.0317
vs.
(2014)
-
@NCKU ()
: ?
97
(N/A 0 !!)
(eg. N/A)
: / (Nearest-Neighbor Estimators)
(2014)
-
@NCKU ()
?
Difficulty in Data Collection
Multiple sources
Different data type/format
Redundant dataset
Difficulty in cross -functional/interdepartmental data collection ()
? ???
()
8am-6pm24hrs
363636
defectsdefects
98
-
@NCKU ()
(problem) (analysis) Shop-floor: defective product, trouble shooting, yield prediction, virtual
metrology, predictive maintenance (PdM), etc.
Management: demand forecast, price prediction, market share, etc.
The complexity of manufacturing significantly depends on product-mix.
? trouble shooting based on the same product-mix
tune
robust forecasting based on the different product-mix
MSExrecipey(!?)( or )
(coefficient of independent variable)
99
-
@NCKU () 100
-
@NCKU ()
A Short Break
Q&A
101
-
@NCKU ()
Cases, Issues and Myths
102
-
@NCKU () 103
Process Diagnosis / Trouble Shooting
()
Yield Prediction
(/)
Feature Selection
()
Process Adjustment
()
-
@NCKU ()
104
-
@NCKU ()
1: Process Diagnosis/ Trouble Shooting ()
AMHS
(cycle time)(sampling)
105
Over 1000 stages!
-
@NCKU ()
106
-
@NCKU ()
1: Process Diagnosis/ Trouble Shooting ()
Wafer Bin Map () Wafer bin maps that show specific spatial patterns can provide clue to
identify process failures in the semiconductor manufacturing.
In practice, most companies rely on experienced for troubleshooting.
However, as IC feature size is continuously shrinking, WBM patterns
become complicated and thus human judgments become inconsistent
and unreliable.
In the semiconductor fabrication process, the circuit probe (CP) test is
used to detect specific failures of each die and thus indicate the test
results with the corresponding bin values.
107
(Liu and Chien, 2013)
-
@NCKU ()
1: Process Diagnosis/ Trouble Shooting ()
108
(Liu and Chien, 2013)
Root Cause
-
@NCKU () 109
Map clustering for
systematic defects
12 defect patterns
and build WBM bank
Decision tree for root
cause classification
(Hsu and Chien, 2007)
1: Process Diagnosis/ Trouble Shooting ()
-
@NCKU ()
110
-
@NCKU ()
2: Feature Selection ()
Feature Selection
In a large-scale dataset, select a few important variables (i.e. columns)
In particular, # of variables (p) much larger than # of observations (n),
i.e. p >> n issue
Address Curse of Dimensionality
The number of observations required exponentially grows to estimate the
function or model parameters.
111
A n
p
X n
k k
-
@NCKU ()
2: Feature Selection ()
Objective (Guyon and Elisseeff, 2003)
improving the prediction performance of the predictors
providing faster and more cost-effective predictors
providing a better understanding of the underlying process that
generated the data. (eg. for process monitoring)
Types
Variable selection: select the best subset of the existing
variables/features without a transformation.
Supervised Learning () with Y as label
Eg. stepwise regression, LASSO, random forest, etc.
Feature extraction (variable transformation): transforming the existing
variables into a lower dimensional space
Unsupervised Learning () with only X
Eg. independent component analysis (ICA), principal component analysis
(PCA), etc.
112
-
@NCKU ()
2: Feature Selection ()
MECE Feature Selection (Taiwan Patent: I564740)
claims that the important variables should not contain overlapping
information and should provide sufficient information.
113
Lee and Chen (2017)
-
@NCKU ()
2: Feature Selection ()
114
Prepared
Dataset
Feature
Selection
LASSO SVM Stepwise
Regression
Voting
by Sampling or K-fold Cross-Validation
Pivot
Analysis
Engineering
Experience
Feature
Validation
Boosting Random
Forest PCA Random
Sampling
Data
Reduction
(p>>n)
Imbalance?
Correlation
Coefficient
Independence
Test
Data
Preprocessing
Yes
No
Knowledge
Management (KM)
Documentation
Tsai and Lee (2017)
-
@NCKU ()
2: Feature Selection ()
Data Reduction (Tsai and Lee (2017)
(!) (i.e., main effect)
xy
z (x->z->y)zyx
(counter)z
: ()
: x ()
115
-
@NCKU ()
2: Feature Selection ()
Voting () When using several selection methods, calculate the selecting
frequency for robust variable selection.
116
SVID Stepwise LASSO Random
Forest Boosting Voting
SVID_003 4
SVID_101 3
SVID_021 3
SVID_040 3
SVID_002 3
SVID_128 2
SVID_062 2
SVID_077 2
-
@NCKU ()
2: Feature Selection ()
K-fold Cross Validation
eg. 10-fold cross validation
117
training fold testing fold
1st iter.
2nd iter.
3rd iter.
10th iter.
Error1
Error2
Error3
Error10
Minimize Error=1
10 Error10=1
-
@NCKU ()
2: Feature Selection ()
118
Class Imbalance
Random sampling deals with the issue. We focus on undersampling
which samples a subset of the majority class.
The main deficiency is that many majority class examples are ignored.
Thus, we sample several subset from the majority class (resampling).
Example
For Y label, vs. = 1000 : 50
Samples 50 at a time for model training
# of replications: 20 times
Rank the variables by the voting
SVID Voting by
Undersampling
SVID_003 19
SVID_101 18
SVID_021 18
SVID_040 18
SVID_002 17
SVID_128 17
SVID_062 17
SVID_077 17
-
@NCKU ()
2: Feature Selection ()
Pivot Analysis ()
119
SVID _Avg _avg Range/Avg.
SVID_003 0.9487 0.2583 1.1439
SVID_101 0.2078 0.5434 0.8934
SVID_021 0.4105 0.8377 0.6845
SVID_040 52.08 71.69 0.3170
SVID_002 2.7256 2.1025 0.2581
SVID_128 8.0523 8.5336 0.0580
SVID_062 0.9430 0.9747 0.0330
SVID_077 1569 1603 0.0216
Mach_ID FAIL TOTAL RATIO
Photo_06 12 384 0.0313
Etch_10 3 295 0.0102
Photo_32 10 1011 0.0099
Photo_17 4 410 0.0098
PVD_02 1 123 0.0081
CVD_14 9 3456 0.0026
Diff_09 1 495 0.0020
Etch_07 5 2769 0.0018
1!
.
-
@NCKU ()
2: Feature Selection ()
Engineering Experience Validation
Y
120
BuzzHand (2014)http://www.buzzhand.com/post_118913.html
-
@NCKU ()
2: Feature Selection ()
Engineering Experience Validation
30
3~4
: : SVID_051SVID_064 38
: SVID_034SVID_082 11
: SVID_115 1
121
3~4
-
@NCKU ()
2: Feature Selection ()
(KM)
122 122
-
@NCKU ()
()
?
123
-
@NCKU ()
/
124
-
@NCKU ()
3: Yield Prediction (/)
(recipe)(Y)eg.
? /
(thickness)(Critical Dimension, CD)(Overlay)(defects)(area)(layer), (yield)
Benefits
(predictive maintenance)
125
-
@NCKU ()
3: Yield Prediction (/)
126
Dataset with
selected variable
Prediction/
Classification
Model
PLS BPNN CNN
K-fold Cross-Validation
R-Squared
MSE
Confusion
Matrix
Model
Evaluation
Boosting Decision
Tree SVR
Random
Sampling
Data
Transformation
Imbalance?
Categorical to
Dummy Var. Z-scores
Feature
Selection
Yes
No
Knowledge
Management (KM)
Documentation
Hung, Lin and Lee (2017)
Data
Discretization
Data Partition (training vs. testing)
-
@NCKU ()
3: Yield Prediction (/)
Testing Dataset with R2, Adjusted R2, Mean Squared Error(MSE), etc.
127
Yie
ld /
Pre
cis
ion /
Metr
olo
gy V
alu
e
Lot
Hung, Lin and Lee (2017)
-
@NCKU ()
3: Yield Prediction (/)
Statistical Control Chart for Classification
128
UCL
LCL
Yie
ld /
Pre
cis
ion /
Metr
olo
gy V
alu
e
Lot
-
@NCKU ()
- Confusion Matrix
FNFPTNTP
TNTP
Accuracy
3: Yield Prediction (/)
1 () 2 ()
1 ()
TP (true positive)
FN (false negative)
(Type II error)
(miss)
2 ()
FP (false positive)
(Type I error)
(false alarm)
TN (true negative)
129
-
@NCKU ()
3: Yield Prediction (/)
(Sensitivity, Recall) 1()
Miss () Rate = 1 Sensitivity
(Specificity) 2()
False Alarm () Rate = 1 Specificity
130
TPSensitivity
TP+FN
TNSpecificity
TN+FP
1 2
1 TP FN
2 FP TN
-
@NCKU ()
3: Yield Prediction (/)
1()
1() (Type I error)
FP Rate = 1 Specificity
TP Rate
= Sensitivity
Model A
Model B
ROC Receiver Operating
Characteristic curve
FP RateTP Rate
TP rateFP rate
TP RateFP RateTrade-off ()
!
131
-
@NCKU ()
3: Yield Prediction (/)
132
Testing
Accuracy TP Rate TN Rate AUC
BPN 71.9% 89.7% 51.7% 70.2%
PLS 78.1% 69.1% 88.3% 78.9%
Model Performance (an example)
BPN
Bad Good
Bad 61 7
Good 29 31
PLS
Bad Good
Bad 47 21
Good 7 53
AUC: Area under the Curve of ROC
-
@NCKU ()
?
133
-
@NCKU ()
134
-
@NCKU ()
4: Process Adjustment ()
Process Adjustment
Manipulate the compensating variables of a process to achieve the
desired process behavior (eg. output close to a target)
135
Time
Y
UCL
LCL
Time
Y
UCL
LCL
-
@NCKU () 136
Shooting Problem
4: Process Adjustment ()
Step1x?
Step2x?
-
@NCKU ()
4: Process Adjustment ()
Step1: x? + (Pivot Analysis)
YX1,,X50
137
Time
Lot ID
Yield
-
@NCKU ()
4: Process Adjustment ()
Pivot Analysis ()
138
-
@NCKU ()
4: Process Adjustment ()
X1X2?
Simple ratio? X1: 5.67/5.33 = 1.0638 , X2: 460/410=1.1220
Range/Avg? X1: 0.33/5.5 = 0.06, X2: 50/435 = 0.1149
()
!? One-way analysis of variance (ANOVA)
!?
139
Row Labels Average of
X1
Average of
X2
0 5.33 410
1 5.67 460
Grand Total 5.50 435
-
@NCKU ()
- 123
p
http://datasci.tw/event/engineer_statistics_170114/
140
(Ling-Chieh Kung)
!!
-
@NCKU ()
4: Process Adjustment ()
Step2: x? Exponentially Weighted Moving Average(EWMA) controller
(Montgomery, 2013)
Assume: a process output gradually goes up and away from the target ()
is widely used to compensate for process shift and noise.
where is a constant (called process gain) like regression coefficient.
+1 is a disturbance and usually autocorrelated.
where = is the prediction error at time t and is weighting factor for EWMA.
This type of process adjustment scheme is called integral control.
141
+1 = + +1
+1 = + 1 = +
= 1 = 1
( +1 ) =
( )
-
@NCKU ()
4: Process Adjustment ()
An Example of Integral Control (Montgomery, 2013)
Figure shows 100 observations on the number average molecular
weight of a polymer, taken every four hours.
the target value T = 2000
The drifting behavior of the process is likely caused by unknown and
uncontrollable disturbances in the incoming material feedstock and
other inertial forces, but it can be compensated for by making
adjustments to the setpoint of the catalyst feed rate x.
142
-
@NCKU ()
4: Process Adjustment ()
An Example of Integral Control (Montgomery, 2013)
Now, we build a equation for the adjusted process
We will forecast the disturbances with an EWMA having = 0.2.
143
+1 2000 = 1.2 + +1
= 1 =
+1 =
1
6 2000
-
@NCKU ()
4: Process Adjustment ()
Temperature less than 588.5
Temperature great than 589.5
144
Lot #
Over time
Lot #
Over time
Thic
kness
Thic
kness
Tsai and Lee (2017)
3.7
2 3
.75
3
.78
3
.81
3
.72
3
.75
3
.78
3
.81
Decrease x
(temperature)
by 1.5
-
@NCKU ()
Multivariate Control Limits
The information is found in the
correlation pattern - not in the
individual variables!
4: Process Adjustment ()
Chang (2012), IBM
The Need for Multivariate Data Analysis - Correlation
145
-
@NCKU ()
?
146
-
@NCKU ()
?
!
()
()
147
-
@NCKU ()
148
-
@NCKU ()
149
R&D
(2015)
-
@NCKU ()
2017/02/11, 2pm3 ()
A3
= () / ()
eg. (56+160+138 mins)/(180+180+180 mins)
=65.6%
150
()
()
()
()14:00-14:11 11 11
14:11-14:16 5
14:16-14:27 11
14:27-14:30 3
14:30-14:43 13
14:43-14:50 7
14:50-14:54 4
14:54-15:00 6
15:00-15:15 15
15:15-15:28 13
15:28-15:33 5
15:33-15:43 10
15:43-15:48 5
15:48-16:01 13
16:01-16:10 9
16:10-16:17 7
16:17-16:22 5
16:22-16:31 9
16:31-16:40 9 9
16:40-16:47 7
16:47-16:52 5
16:52-16:55 3
16:55-17:00 5
43
140
20
88
42
50
(B)
(D)
(C)
13
124
BA C
(A)
(B)
(C)
-
@NCKU ()
80% 85%
(80% 85%)
200 * 80% = 160 () 200 * 85% = 170 ()
(Dashboard)160Run ()
(eg. )
151
-
@NCKU ()
=
=/
/
=
=
(Labor Eff)
1(Machine Util)
=0.8
10.656= 2.33 ()
152
-
@NCKU ()
( / time-motion study)
()
(/)
()
~
( )
153
-
@NCKU ()
?
?
?
(robust)?
Training datasettesting dataset?
?
?
?
(re-train)?
?
?
?
..
154
-
@NCKU ()
WIP?
?
throughput?
?
?
?
target?
product-mix?
? ?
Continuous Delivery??
?
?
..
155
?
-
@NCKU ()
!!
(Dec. 31, 2016) "2016"
156
-
@NCKU ()
A Short Break
Q&A
157
-
@NCKU ()
From Predictive to Prescriptive
158
-
@NCKU ()
159
?
?
?
-
@NCKU ()
160
-
@NCKU ()
161
http://thewhen.pixnet.net/blog/post/42532195-
%E5%A1%94%E7%BE%85%E5%8D%A0%E5%8D%9C%3A%E6%88%91
%E8%A9%B2%E5%A6%82%E4%BD%95%E5%A2%9E%E5%8A%A0%E8
%B2%A1%E9%81%8B%3F
http://www.zou1.com/news/show-97765.html
-
@NCKU ()
162
-
@NCKU ()
163
http://www.ettoday.net/news/20150924/569618.htm
-
@NCKU () 164
-
@NCKU ()
.
XD
165
-
@NCKU ()
()
!
166
-
@NCKU ()
167
-
@NCKU ()
(Decision under Certainty)
(Decision under Risk)
(Decision under Strict Uncertainty)
168
-
@NCKU ()
169
1. () 2. ()
1. () 2. () 3. !! ()
-
@NCKU ()
170
(Decision under Certainty)
God2GO
(100%)
-
@NCKU ()
(Decision under Risk)
171
0.2
0.8
()
Weather2GO
0.2 +0.8 (!?)
-
1/6, 3.5
3.5 (!?)
-
@NCKU ()
(Decision under Risk)
172
0.2
0.8
()
Weather2GO
$0 $100 ()
$200 ()
$0
0.2x0
+0.8x200
=160
0.2x100
+0.8x0
=20
-
@NCKU ()
(Decision under Risk)
173
$0 $100 ()
$200 ()
$0
0.2x0
+0.8x200
=160
0.2x100
+0.8x0
=20
0.8 ()
100,000 ?
-
@NCKU ()
()
()
174
-
@NCKU ()
(Decision under Strict Uncertainty)
175
Based on your
Preference
Structure
1
2~
I2GO
-
@NCKU ()
(Decision under Strict Uncertainty)
50
176
42 21 8
BBQ 35 28 13
KTV 28 25 23
? Revised from Chien (2005)
-
@NCKU ()
(Decision under Strict Uncertainty)
(The Maximin Payoff Criterion)
177
Min
Return
42 21 8 8
BBQ 35 28 13 13
KTV 28 25 23 23
-
@NCKU ()
(Decision under Strict Uncertainty)
(The Maximax Payoff Criterion)
178
Max
Return
42 21 8 42
BBQ 35 28 13 35
KTV 28 25 23 28
-
@NCKU ()
(Decision under Strict Uncertainty)
(The Minimax Regret Criterion) (Savage, 1951)
Step 1
Step 2(minimax)
179
Max
Regret
0 7 15 15
7 0 10 10
14 3 0 14
42 21 8
BBQ 35 28 13
KTV 28 25 23
-
@NCKU ()
(Decision under Strict Uncertainty)
(preference structure)
180
42 21 8
BBQ 35 28 13
KTV 28 25 23
?
-
@NCKU () 181
Capacity Planning Problem (Newsboy problem over time)
Capacity Plan: A, B, C or D (demand)?
Two Types of Risk
Capacity Shortage : loss of sales, loss of market share
Capacity Surplus : machine idleness, inventory, holding cost
Time
Demand
Demand
A
B
C
-
@NCKU ()
Newsboy problem
0 ()
: ?
182
-
@NCKU () 183
Capacity Planning Problem (more demand scenarios)
Two demand forecasts: Marketing vs. Sales
Should capacity level follow marketing, or sales, or ?
How about more than two demand scenarios?
Time
Demand
Marketing Forecast
Sales Forecast
0
Capacity Level?
-
@NCKU ()
Empirical Study (Lee and Chiang, 2016)
Empirical Study- Taiwan TFT-LCD Manufacturer
We introduce a two-phase framework to decide the best capacity level
reducing the losses (i.e., regret) of capacity surplus and capacity
shortage.
1st stage: Demand Forecast () Linear Regression
Autoregression
Neural Network
2nd stage: Capacity Decision () Expected Value (EV, )
Minimax Regret (MMR, )
Stochastic Programming (SP, )
184
-
@NCKU ()
Empirical Study
2-stage approach
Demand Forecast
Capacity Decision
VDGP
LI: latent information
(Chang et al., 2014)
185
Lee and Chiang (2016)
-
@NCKU ()
Empirical Study
Capacity Regret Assessment
Analytic Hierarchical Process (, AHP) (Satty, 1980)
186
Capacity Regret
Assessment
CustomerInventory
Market
Share
Profit
Loss
Loss of
Falling Price
Inventory
Loss
Holding
CostReputation
Lower
Capacity
Level
Higher
Capacity
Level
ALTERNATIVE
OBJECTIVE
ATTRIBUTE
Lee and Chiang (2016)
-
@NCKU ()
Empirical Study
A Tradeoff Between Two Risks
187
Capacity Decision
-
@NCKU ()
Empirical Study
Data Source
Nov. 2013 to Oct. 2014
aggregate demand for a specific technology node
Dataset: training data (9 months) and validation data (3 months)
Objective: determine the capacity level for the future 3 months (Aug.
2014 to Oct. 2014) based on historical demand until July 2014.
188
0
10,000
20,000
30,000
Training
Dataset
Validation
Dataset
Technology
Node A
Time
Demand
Lee and Chiang (2016)
-
@NCKU ()
0
10,000
20,000
30,000
Training
Dataset
Validation
Dataset
Technology
Node A
Time
Demand
Empirical Study
189
1st phase: Demand Forecast
linear regression (OLS), autoregression (AR), neural network (NN)
Criterion: Mean Squared Error (MSE): NN < AR < OLS
Linear Regression
Autoregression
Neural Network
-
@NCKU ()
Empirical Study
2nd phase: Capacity Decision
AHP gives pair-wise comparisons of the capacity surplus and shortage,
and calculate the eigenvector () as weights of two risks.
31 individuals with TFT-LCD industrial backgrounds
2 for 0-2 yrs, 9 for 2-5 yrs, 13 for 6-9 yrs, 7 for more than 10 yrs of service
one third have undergraduate degrees and the others for graduate degrees
= 0.521 and = 0.479.
190
*Current policy builds up the capacity level based on the maximal forecast demand.
**Avg. Reg.: the weighted sum of squared error (WSSE) over the number of validation periods
Month Actual
Demand
Capacity Decision Current Policy
EV MMR SP
Aug. 2014 19499 20207 16941 17157 18922 Sep. 2014 18110 21305 17921 18306 19135 Oct. 2014 15059 22404 18536 19291 18885
Standard Deviation 2271.3 1098.2 804.7 1068.1 134.8
Average Regret - 1796.0 1032.2 1113.8 874.8
-
@NCKU ()
Empirical Study
191
0
500
1000
1500
2000
0 300 600 900 1200
StandardDeviation
AverageRegret
Current
Policy
EVMMR
SP
Poor
Better
Better Poor
Lee and Chiang (2016)
-
@NCKU ()
(scenarios)
Butwhat is the next step?
100%
A95%
B90%
AB?
192
-
@NCKU ()
~
193
Testing
Accuracy TP Rate TN Rate AUC
BPN 71.9% 89.7% 51.7% 70.2%
PLS 78.1% 69.1% 88.3% 78.9%
3: Yield Prediction (/) Classification Confusion Matrixtrade-off
type-I and type II
BPN
Bad Good
Bad 61 7
Good 29 31
PLS
Bad Good
Bad 47 21
Good 7 53
AUC: Area under the Curve of ROC
-
@NCKU ()
?
194
-
@NCKU ()
195
170cm
165cm 175cm
172cm
-
@NCKU ()
.
196
-
@NCKU ()
(! !!)
()
~ ()
Manufacturing Hand Make ~
()
()
(scenarios)
()
()
197
-
@NCKU ()
- (tornado diagram) Adjustment range: 10%
198
UCL LCL
increase
decrease
SVID_003
SVID_101
SVID_021
SVID_040
SVID_002
SVID_128
Variable
Y or Metrology
-
@NCKU ()
(Analytics)?
Hillier and Lieberman
Analytics is the scientific process of transforming data into insight for
making better decisions
Descriptive analytics () Using innovative techniques to locate the relevant data and identify the
interesting patterns in order to better describe and understand what is
going on now.
Predictive analytics () Using the data to predict what will happen in the future.
Prescriptive analytics () Using the data to prescribe what should be done in the future. The
powerful optimization techniques of operations research are what are
used here.
199
-
@NCKU ()
200
Source: CRISIL GR&A analysis
Operations Research ()
-
@NCKU ()
vs.
201
Yield
Time
Cost
Yield
New Tape-out (NTO)
/
Device
optimization
during process
integration
Engineering
Parameter
Optimization
Tool Matching
Equipment Health
Monitoring (EHM)
Spec Setting/ SPC
Virtual Metrology
R2R Control/ FDC
PM/PdM
Virtual Material Quality Investigation
Yield ramp
during pilot
production
Process turning
during volume
production
-
@NCKU ()
?
() ()
202
http://www.supergameasset.com/male-warrior-game-asset.html
https://68.media.tumblr.com/618419c9c129cdd1176855d202334e25/tumblr_o829ynvbH61udl64oo1_400.gif
VS
-
@NCKU ()
The Wyndor Glass Co. products high-quality glass products,
including windows and glass doors. (Hillier and Lieberman, 2010)
It has three plants. Plant 1 makes aluminum frames and hardware
Plant 2 makes wood frames
Plant 3 produces the glass and assembles the products
Top management has decided to revamp the product line.
Production capacity are released to launch
two new products having large sales potential:
Product 1: An 8-foot glass door with aluminum framing
Product 2: A 46 foot double-hung wood-framed window
203
?
http://img.archiexpo.com/images_ae/photo-g/55456-4334799.jpg
https://www.andersenwindows.com/
-
@NCKU ()
This problem can be recognized as LP problem of the classic product mix () type.
Linear Programming (, LP)
Formulation as a Linear Programming Problem
x1: number of batches of product 1 produced per week
x2: number of batches of product 2 produced per week
Z: total profit per week from producing these two products
204
?
Production Time Available
Time Plant Product 1 Product 2
1 1 0 4
2 0 2 12
3 3 2 18
Profit ($1,000) $3 $5 0,
1823
122
4 s.t.
53
21
21
2
1
21
xx
xx
x
x
xxZ
-
@NCKU () 205
Graphical Solution
The resulting region of permissible values of (x1, x2), called the
feasible region.
?
0,
1823
122
4 s.t.
53
21
21
2
1
21
xx
xx
x
x
xxZ
-
@NCKU () 206
Graphical Solution
Move the objective function with fixed slope through the
feasible region in the direction of improving Z.
?
21 (x1) 62 (x2)
!
-
@NCKU ()
?
()
KPI ()
solution
207
-
@NCKU ()
208
DS in
MFG
1. 2. (eg.
) 3. ()
-
@NCKU ()
(Throughput)?
209
...
-
@NCKU ()
210
-
@NCKU ()
Big Data?
Pattern?
211
-
@NCKU ()
?
212
-
@NCKU ()
Market Structure and Game Theory (external analysis)
Productivity driving Price reduction (!?)
Lee and Johnson (2015)
Curse of Productivity!?
213
-
@NCKU ()
The source of complexity in manufacturing is variability (). Automation () is a powerful way to reduce variability.
Big Science facilitates automation and real-time decision.
However
1. (panacea)
2. IT"
3.
4. data ()
5. ~
()().
214
-
@NCKU ()
KPI Hierarchy
215
Dupont
ROE
Profit
Margin
Asset
Turnover
Financial
Leverage
Profit
Sales
Assets
Cost Accounts Receivable
Cash + Inventory +
Bottleneck
Machine
Other
Assets +
Labor Material Floor
Space
Non-bottle.
Machine
Fin
an
cia
l Ind
ex
Ma
nu
factu
ring In
de
x
Reso
urc
e
Decis
ion
TAT = Turn Around Time
CVP = Customer Volume Performance
CIP = Customer (Line) Item Performance
-
@NCKU () 216
(USnews2017)
USNews (2017).
http://money.usnews.com/careers/best-
jobs/operations-research-analyst
-
@NCKU () 217
(2014)
-
@NCKU ()
~
~
~
218
-
@NCKU ()
References 2015
2014
Dong, Z.-H. and C.-Y. Lee, 2017. Equipment Health Monitoring of Analytic Heretical Process Index in Semiconductor
Manufacturing. 27th International Conference on Flexible Automation and Intelligent Manufacturing (FAIM2017), 27-
30 June 2017, Modena, Italy.
Guyon, I. and A. Elisseeff, 2003. An Introduction to Variable and Feature Selection. Journal of Machine Learning
Research, 3, 11571182.
Han, J., M. Kamber, and J. Pei, 2011. Data Mining: Concepts and Techniques, 3rd edition, Morgan Kaufmann ()2008
Hillier, F. S.and G. J. Lieberman, 2010. Introduction to Operations Research, 9th ed., McGraw-Hill, New York. ()
Hsu, S. and C. Chien, 2007. Hybrid Data Mining Approach for Pattern Extraction from Wafer Bin Map to Improve
Yield in Semiconductor Manufacturing. International Journal of Production Economics, 107, 88103.
Hung, S.-Y., Y. L. Lin, C.-Y. Lee, 2017. Data Mining for Delamination Diagnosis in the Semiconductor Assembly
Process. 27th International Conference on Flexible Automation and Intelligent Manufacturing (FAIM2017), 27-30
June 2017, Modena, Italy.
Lee, C.-Y. and M.-C. Chiang, 2016. Aggregate Demand Forecast with Small Data and Robust Capacity Decision in
TFT-LCD Manufacturing. Computers & Industrial Engineering, 99, 415422.
Lee, C.-Y. and A. L. Johnson, 2013. "Operational Efficiency", book chapter edited in: Badiru, A. B. (Editor),
Handbook of Industrial and Systems Engineering, 2nd Edition, 17-44, CRC Press.
Lee, C.-Y. and A. L. Johnson, 2015. Measuring Efficiency in Imperfectly Competitive Markets: An Example of
Rational Inefficiency. Journal of Optimization Theory and Applications, 164 (2), 702722.
Liu, C.-W. and C.-F. Chien, 2013. An intelligent system for wafer bin map defect diagnosis: An empirical study for
semiconductor manufacturing. Engineering Applications of Artificial Intelligence, 26, 14791486.
Savage, L. J. 1951. The theory of statistical decision, Journal of the American Statistical Association, 46, 5567
Tsai, T.-L. and C.-Y. Lee, 2017. Data Mining for Yield Improvement of Photo Spacer Process in Color Filter
Manufacturing. 27th International Conference on Flexible Automation and Intelligent Manufacturing (FAIM2017), 27-
30 June 2017, Modena, Italy.
219
-
220
Contact Information:
name: (Chia-Yen Lee) phone: 06-2757575 34223 email: [email protected]
web: https://polab.imis.ncku.edu.tw/
mailto:[email protected]