Download Network Representation of a Supply Chain - LGO

LGO-SDM ALUMNI
On-Line Seminar Series
December 15, 2000
1
Supply Chain Modeling
and Optimization
Stephen C. Graves
MIT, E40-439
Leaders for Manufacturing Program
[email protected]
http://web.mit.edu/sgraves/www/
2
Overview
•
•
•
•
•
•
Supply chain management: definition & issues
Framework for supply chain modeling
Example: safety stock placement
Example: supply chain configuration
Example: inventory hedge for cyclic demand
Other examples and future plans
3
Supply Chain Management
• A process-oriented integrated approach to procuring,
producing and delivering products and services to
customers
• Includes suppliers, internal operations, channel
partners and end customers
• Covers the management of materials, information and
funds
4
Challenges and Issues
•
•
•
•
•
Process orientation, rather than functional
Process metrics and accounting
IT as an enabling technology
Need to cross organizational boundaries
Need to focus on system performance rather than
department or company performance
5
Network Representation of a
Supply Chain
US Factory
US DC
Americas Demand
European DC
European Demand
US Suppliers
Singapore Factory
Off Shore Suppliers
Kit Suppliers
Asia/Pacific DC
Asia/Pacific Demand
6
Framework for Supply
Chain Modeling
• Decision Hierarchy
– Strategic --- network design
– Tactical --- flow planning; countermeasures for
variability and uncertainty
– Operational --- scheduling and control
• My interest – tactical level
7
Strategic Safety Stock
Placement
• Intent - a tactical model to determine the amount and
positioning of safety stocks in supply chains
• Intent - a tactical model to support supply chain
improvement teams
8
KIMES 100
9
KIMES 100
10
KIMES 100
11
Supply Chain BEFORE
12
Supply Chain LEAD TIMES
13
Supply Chain COSTS
14
Supply Chain OPTIMIZED
15
Supply Chain IMPLEMENTED
16
Key Benefits
• Shows value from “holistic” perspective
• Formalizes inventory-related supply chain costs, and
provides an optimal benchmark
• Provides framework and terminology for crossfunctional debate
• Shows the effectiveness of inventory, strategically
positioned in a few places to de-couple the supply
chain
17
Key Learnings
• De-couple supply chain prior to a high-cost added
stage; and prior to product explosion
• Substitute information for inventory
• Postpone product differentiation step
• Win-win from optimizing multi-company supply chain
• Value of a standard terminology and a neutral tool
18
Supply Chain Configuration
for New Products
• How to configure a new product’s supply chain?
• Two vendors can deliver an identical product.
– A quotes 100 days at $1.00 per unit
– B quotes 3 days at $1.10 per unit
• Which one do you pick?
19
Problem Statement
• A supply chain configuration is the set of options
selected for each stage in supply chain
• Stages include procurement; production, assembly
and test processes; distribution channels; and
transportation modes
• Intent: develop a DSS for determining options in SC
configuration, given a stable product design
20
Digital Camera Example
• Pro-summer model
– Monthly demand has a mean of ~550 units and a
standard deviation of ~50
• Three major subassemblies
• Two customer markets: US and export
21
Relevant Supply Chain Costs
•
•
•
•
•
Cost-of-goods sold
Safety stock cost
Pipeline stock cost
Time-to-market cost
Quality cost
22
Current Practice = Target
Costing
• Target for unit manufacturing cost (UMC) set based
on market price and gross margin
• UMC is then allocated to subassemblies
• Designers independently source their portion of the
supply chain
• Other factors (functionality, quality, flexibility)
considered based on thresholds
23
Digital Camera Supply Chain
24
Digital Camera Options
Component/Process
Description
Raw Silicate
Option
Wafer Fab
1
2
1
Wafer Pkg. and Test
1
CCD Assembly
1
Miscellaneous Components
Parts w/ 8 Week LT
1
1
2
3
4
1
2
3
Parts w/ 4 Week LT
Product.
Time
Cost
60
$5
20
$8
30
$800
8
$825
10
$200
5
$225
5
$200
2
$250
30
$200
40
$105
20
$108
10
$109
0
$110
20
$175
10
$177
0
$179
Note: All data has been disguised by scaling
25
Digital Camera Options
Component/Process
Description
Parts w/ 2 Week LT
Parts on Consignment
Circuit Board Assembly
Camera Body
Accessory Processing
Local Accessory Inv.
Camera Assembly
Central Distribution
US Demand
Export Demand
Option
1
2
1
1
2
1
2
1
1
1
2
1
1
2
1
2
Product.
Time
Cost
10
$200
0
$203
0
$225
20
$225
5
$300
70
$650
30
$665
40
$100
10
$60
6
$420
3
$520
5
$180
5
$12
1
$25
11
$15
2
$40
26
Three Solution Approaches
• Minimize unit manufacturing cost
• Minimize production time
• Minimize supply chain costs
27
Solution Comparison
Current
Policy
COGS ($MM)
Min UMC Min Prod
Time
Min SC
Costs
17.8
17.8
19.4
18.0
1.3
1.2
0.6
0.9
Total Configuration Cost
19.1
19.0
20.0
18.9
Unit Manufacturing Cost
$3,756
$3,756
$4,078
$3,794
Length of Longest Path
127 days 127 days 45 days
Inventory Cost ($MM)
118 days
28
Digital Camera Options
Component/Process
Description
Raw Silicate
Wafer Fab
Option
1
2
1
Wafer Pkg. and Test
1
CCD Assembly
1
Miscellaneous Components
Parts w/ 8 Week LT
1
1
2
3
4
1
2
3
Parts w/ 4 Week LT
Product.
Time
60
20
30
8
10
5
5
2
30
40
20
10
0
20
10
0
Cost
$5
$8
$800
$825
$200
$225
$200
$250
$200
$105
$108
$109
$110
$175
$177
$179
29
Digital Camera Options
Component/Process
Description
Parts w/ 2 Week LT
Option
1
2
Parts on Consignment
1
Circuit Board Assembly
1
2
Camera Body
1
2
Accessory Processing
1
Local Accessory Inv.
1
Camera Assembly
1
2
Central Distribution
1
US Demand
1
2
Export Demand
1
2
Product.
Time
10
0
0
20
5
70
30
40
10
6
3
5
5
1
11
2
Cost
$200
$203
$225
$225
$300
$650
$665
$100
$60
$420
$520
$180
$12
$25
$15
$40
30
Role of Holding Cost
Raw Silicate
Wafer Fab
Wafer Pkg. and Test
CCD Assembly
Miscellaneous Components
Parts w/ 8 Week LT
Parts w/ 4 Week LT
Parts w/ 2 Week LT
Parts on Consignment
Circuit Board Assembly
Base Assembly
Accessory Processing
Local Accessory Inv.
Digital Capture Device Assembly
Central Distribution
US Demand
Export Demand
15%
1
1
1
1
1
1
1
1
1
1
2
1
1
1
1
1
1
30%
1
1
1
1
1
3
2
1
1
1
2
1
1
1
1
2
2
45%
1
1
1
1
1
4
3
2
1
1
2
1
1
1
1
2
2
60%
1
2
1
1
1
4
3
2
1
1
2
1
1
1
1
2
2
31
Inventory Investment and
UMC Interaction
Initial Investment
($MM)
UMC
($/unit)
COGS
($MM)
3.3
3,773
17.9
2.8
3,794
18.0
2.7
3,800
18.1
2.5
3,825
18.2
32
Key Learnings
• SC optimization saves three times the savings from
SIP
• Optimization did not make some “obvious” choices
• Increasing unit manufacturing cost by $37 is
significant.
• As you move farther downstream in the supply chain,
higher cost options can be more attractive
• More complex the supply chain, more likely
optimization will find opportunities
33
Next Steps
•
•
•
•
Verify/validate the model in practice
Software to disseminate expected, March 01
Incorporate side constraints, e.g. number of vendors
Extend to consider capacity options
34
Creating an Inventory
Hedge for Cyclic Demand
• How to plan long lead-time, custom product subject to
cyclic demand?
• Motivations: Teradyne case
• Intent: demonstrate effectiveness of planning tactic
(the hedging policy) and develop a stylized model for
guiding its deployment
35
Teradyne Case
• Flagship product is Catalyst, for testing linear and
mixed signal devices
• Tester sells for $1.5 - 2.0 M
• Each sale is for a customized product
• Product is complex, 100’s of PCB’s, 10000’s of
components
• Customer delivery time << Manufacturing lead time
• Demand is volatile with limited predictability
36
Product Structure
• Three levels to BOM: option level, PCB, component
• 175 options; each option consists of 1 - 8 PCB’s
• Each tester consists of about 50 options, plus work
station, test head, mechanical assembly
37
Demand
• Aggregate demand very volatile, subject to bull whip,
alternates between up cycle and down cycle
• About 30% of options have stable demand; and
account for over 70% of material cost
• 50% of options are used infrequently, account for less
than 15% of cost
38
Quarterly Sales Data (Q2 1994 - Q1 1999)
3
Sales
2.5
2
1.5
1
0.5
0
Quarter
Figure 1: ICD’s Sales by quarter for past 5 years
39
Production Planning
• Testers are assembled to order
• MPS at the option level to fill material pipeline
• MPS assumes an aggregate demand rate (x per
week)
• MPS assumes a planning bill to plan open testers
• Scheduling group matches ‘potential’ and booked
orders to open testers, and re-schedules options as
necessary
• Options not in planning bill are scheduled in ad hoc
way
40
Hedging Policy
• MPS for next L weeks assumes current aggregate
demand rate, x per week.
• MPS beyond L weeks assumes a “maximal” demand
rate, y per week, y > x
• This creates an intermediate-decoupling inventory,
sized to meet a “maximal” demand rate
• This inventory permits ICD to ramp quickly when
transition from down to up cycle
41
100%
90%
80%
Possible Hedging Points
Cost
70%
60%
50%
40%
30%
20%
10%
0%
Lead-Time
Cost Accrual Function -- Suggests hedging points
42
Implementation Results
• Considerably better performance: $50 - $100 million
in incremental revenue was realized over recent
upturn
• More systematic treatment of materials management
than before
43
Research Questions
• The role of hedging stocks in environments subject to
non-stationary demand
– Where does one locate the hedging stock?
– How is the hedging stock sized?
– What kind of service performance will be attained
for a given hedging stock configuration
44
Hedging Model
• Principal issues to address
– Need an effective strategy to manage the inventory
pipeline when demand is non-stationary and
changes without predictability
– Need model to characterize inventory levels and
service performance, and illuminate tradeoffs
45
Diagram of a supply pipeline
Replenishment orders
t=m
System ships
t=0
Cumulative value of material purchased as a function of time
46
m-L
L
Intermediate-decoupling Inventory
FGI
• Where to locate the decoupling stock
(how to set L)?
– decoupling stock is sized for the
high-demand rate
• How does one characterize service
level, given non-stationary demand?
47
Optimization Problem
• Decision variables L,SD,SH
• Objective function: minimize average expected
holding costs over a cycle
• Subject to: Target fill rates during the low to high
transient sub-cycle, the high demand and low
demand sub-cycles
48
Minimize
 E [OH ]
Subject To :
(S D - lD L)
lD L
(S H - lH L)
lH L
tD
tH
( m - L ) ( S H - l H L ) L  2  - ( L l H ) 3 / 2 + ( L l D ) 3 / 2 + 3 S D ( L l H )1 / 2 - 3 S D ( L l D )1 / 2
+  
 t D H
lH L
m
m3
(l H - l D ) L
S H ,S D 0
lLm
49
140000
120000
TC*
100000
80000
60000
40000
20000
0
0
5
10
15
20
25
30
35
L
Figure 2: The optimal objective function value as a function of L
50
Conclusions and Learnings
• Can use model to understand the value of hedging
policy
• Can use optimization model to determine the location
of a decoupling inventory and base stock levels for
both inventories
• Find that optimal value of L lies near the kink in the
lead-time cost accrual profile
51
Other Examples
• Seasonal planning with uncertainty (Monsanto)
• Evaluation of flexibility benefits (GM)
52
Current/Future Plans
• Planning tactics for low-volume custom-engineered
products (ABB)
• Evaluation of capacity options in supply chain design
(TBD)
• Evaluation and utilization of POS data (Kodak?)
53
Further Information
• Papers and theses:
http://web.mit.edu/sgraves/www/
• SIP model software:
http://web.mit.edu/lgorg3/www/
• [email protected]
617.253.6602
54