Transcript Simulation model of a mixed Make-to-Order and Make

Simulation-based Optimization
for Supply Chain Design
INRIA Team
April 7, 2004
Torino-Italy
Keys issues in supply chain design

Uncertainties and risks




Interrelation between decisions at different levels



Strategic decisions
Operational decisions
Multiobjective


Demand fluctuation
Supply disruption
Transportation instability
Costs vs. Customer service level
Characteristics of the case studies



Demand seasonality, unstable transportation lead-time
Supplier selection, inventory control
Cost, lead-time, demand fill-rate
2
A case study from textile industry
(actual situation)

Company outsources its production to outside contractors and
focuses only on product design, marketing and distribution issues,

One part of the global supply chain of the company, which
distributes a single type of product “classic boot” around Europe, is
considered,

According to the inventory control policy,
the DC places replenishment orders periodically,

A unique supplier in Far East is employed
for stock replenishment,

There is only one transportation link
that connects the DC and the supplier,

After a period of supply lead-time, required boots are collected into
containers and transported by boat from Far East to a European
harbor and then to the DC by trucks
3
A case study from textile industry
(evaluated scenario)
Company motivations
1. Current order-to-delivery lead-time (period from the moment
when the DC places an order to the moment when the DC receives
required products) is relatively long:
“long distance (Far East-Europe)+boat as the principle carrier”
2. High variability demands for “classic boot” + frequently stock-out
Actual
Cheapest
Normal
Fastest
4
Problem

Optimal supply portfolio



Possibly multi-supplier
Combinations of various transportation
modes
Traditional approaches



Analytical Hierarchic Process (AHP)
Elimination
Mathematical programming
5
Why simulation-optimization?

Strategic + operational decisions




Dynamic in nature



Supply chain network design
Order assignment ratio
Inventory control parameters
Demand seasonality
Unstable transportation time
Original
work !
Multiple criteria


Total costs
Backlog ratio
6
The proposed methodology
Objective: To design supply chain networks
that are efficient in real-life conditions
Optimizer
Supply chain
configurations
Performances
estimations
Solution Evaluator
7
Key requirements

Optimizer




Combinatorial optimization
Capable to learn from previous evaluations
Suitable for multiobjective optimization
Evaluator



Faithful and efficient evaluation
Capable to catch stochastic facts
Flexible for different SC structures
Genetic
Algorithm
Rule-based
Simulation
8
What is Genetic Algorithm?



A search algorithm
Large and non-linear search space
Based on the mechanics of natural
selection and evolution

Generation by generation
Selection
Crossover

Mutation


9
Characteristics of GA

Probabilistic in nature

Search from one population to another


Use only objective function information
to guide the search direction
Need a sufficient number of simulation
runs, time-consuming
10
An example
Chromosome
Phenotype
 Integer value
 Network configuration
 Schedule
…
Gene
Replenishment level: 1*27+0*26+1*25+0*24+0*23+0*22+1*21+1*20 = 163
Network configuration:
Supplier1 Supplier3 Supplier7 Supplier8
11
Simulation-based optimization
Step1: Generate an initial population of chromosomes
1 chromosome = 1 network configuration
Boat +truck
0 1 1 0
0 0 1 0
0 1 0 0
0 0 1 1
Plane + truck
Supplier 1
Far East
Boat + plane + truck
Boat +truck
Plane + truck
Supplier 2 Boat + plane + truck
Far East
Truck Truck
Delivery
Distribution
Center
European
Market
Supplier 3
Europe Supplier 4
Europe
1 0 0 0
12
Simulation-based optimization
Step1: Generate an initial population of chromosomes
1 chromosome = 1 network configuration and 1 set of parameters
Step2: Evaluate all chromosomes by simulation
Fitness = f (KPI1, KPI2, …)
Boat +truck
Plane + truck
Supplier 1
Far East
Boat + plane + truck
Boat +truck
Plane + truck
Supplier 2 Boat + plane + truck
Far East
Truck Truck
Supplier 3
Europe Supplier 4
Europe
Delivery
KPI
Distribution
Center
European
Market
 Purchasing cost
 Transportation cost
 Inventory cost
 Unmet demand
13
Simulation-based optimization
Step1: Generate an initial population of chromosomes
1 chromosome = 1 network configuration
Step2: Evaluate all chromosomes by simulation
Fitness = f (KPI1, KPI2, …)
Step3: Selection of chromosomes for crossover
Step4: Produce offspring by crossover and mutation
Step5: Repair of offspring for feasibility
Return to Step2
14
Application on the case study
Customer
Demand quantity*
Demand interval*
Behavior type
Expected leadtime
Service priority
Distribution Center
Supplier
FOB Price*
Storage capacity
Duties*
Over-capacity cost
Supply leadtime*
Holding cost
Minimum order size*
Ordering cost
Engagement cost
Transportation Link
Transportation leadtime*
Carrier capacity
Unit shipment cost*
Batch shipment cost*
15
GA specifications in SGA case
Binary variables
for supplier selection decisions





Integer variables
for assignment weights
A integer variable
for replenishment level
Population size = 12
Generation number = 500
pCrossover = 0.9
pMutation = 0.01
Fitness = Purchasing costs + Transportation costs
+ Inventory costs + ß*Backlogged
ß (€/pair) : punishment factor
16
Principal assumptions


Simulation horizon = 3 years
Customer behavior



Inventory control policy



Non-patient customer
Weekly demand: N( 783, 100 )
Periodic replenishment
Replenish period = 7 days
Proportional order assignment
17
Single-objective GA (SGA)
Minimize the total costs
Total costs = Cpurch. + Ctrans. + Cinventory + Clost sales
Best-so-far solution:


1- Unique supplier from Far East: Supplier B
2- Two transportation links :
Boat + truck (73.7%) and Plane + truck (26.3%)
3- Replenishment level: 10800
4- Total costs: 1.48 e+006 €
7,9% Transportation
Total Costs €
2,10E+06
1,6%
1,90E+06
Inventory
9,6% Lost sales
1,70E+06
1,50E+06
1,30E+06
0
25
50
75 100
Generation
125
150
Purchasing
80,9%
18
GA specifications in MOGA case
Binary variables
for supplier selection




Integer variables
Integer variables
for transportation
for order quantity
allocation weights Integer variables
for reorder point
Population size = 100
Generation number = 2000
pCrossover = 0.9
pMutation = 0.1
19
Principal assumptions in MOGA



Simulation horizon = 4 years
Simulation replications = 10 times
Customer behavior



Non-patient customer
Weekly demand: N( 783, 100 )
Inventory control policy


(R, Q)
Replenish period = 7 days
20
Multi-objective GA (MOGA)

Modifications regarding to SGA


Pareto optimality; Fitness assignment; Solution filter
Two objectives


Minimize the total cost
Maximize the demand fill-rate
Demand Fill-Rate
100%
95%
90%
85%
80%
8
9 10 11 12 13 14 15 16 17 18
Unit Cost (€)
21
Innovations of the proposed approach

Capable to optimize both
 supply chain configurations
 operational decisions

Uncertainties and risks covered

Multi-objective decision-making
22