Integrated Supply Chain

1

• Integrated Supply Chain
• Analysis and Decision Support(I98-S01)

2
RESEARCH TEAM
Textiles and Apparel Mgmt. Industrial
Engineering Industrial Engineering Textiles and
Apparel Mgmt. Industrial Engineering Industrial
Engineering
INVESTIGATORS G. Berkstresser S. Fang R. King T.
Little H. Nuttle J. Wilson
3
RESEARCH TEAM
STUDENTS S-H. Chen Y. Liao A. Medaglia
Ph.D. Operations Research Ph.D. Industrial
Engineering Ph.D. Operations Research
4
Background

• Supply chains involve the activity and
  interaction of many entities.
• Successful operation requires coordination of
  decision making among the entities.
• Decisions must be made in settings involving
  vagueness and uncertainty.
• Performance evaluation is complicated by the
  presence of conflicting objectives.
• These issues become more serious as the number of
  operations and number of players in the chain
  increase.

5
Background (cont.)

• Performance measures, such as service level and
  cost, and system parameters, such as inventory
  levels, plant capacities, and leadtimes, are
  understood in a general sort of way.
• Precise relationships between system parameters
  and performance measures are really not known
  and, in fact, will change from one time to
  another depending on uncertain factors such as
  customer demand and manufacturing yields.

6
Background (cont.)

• Fuzzy mathematics permits one to directly model
  imprecise relationships using linguistic
  variables.
• While fuzzy logic permits one to do approximate
  reasoning to obtain useful results.

7
Objectives

• Attack Critical Soft Goods Supply Chain
  integration and decision support problems using
  Fuzzy Mathematics and Neural Network technologies

• Develop the capability to model soft goods supply
  chain design and decision making problems using
  this framework.
• Develop mathematical models for specific
  scenarios involving both numerical and linguistic
  data.
• Design and evaluate approaches for solving the
  models.
• Prototype a decision support system.

8
Multi-Customer Due-Date Bargainer
9
Multi-Customer Due-Date Bargainer
Multi-Customer Due-Date Bargainer (MCDDB)
is a new tool for due-date negotiation between a
manufacturer and customers
Combines the Order Management,
Resource Management, Due-Date Bargaining and
Schedule Management Function into one package.
MCDDB
10
Steps of MCDDB

• Step I

Input order data.

• Step II

Calculate the Manufacturers prefer Due-Date
(no overtime).

• Step III

Calculate the Fuzzy Promised Due-Date balancing
overtime use and delayed delivery.

• Step IV

Execute bargaining process with dissatisfied
customers.
11
Source MCDDB (Beta Version)
12
Source MCDDB (Beta Version)
13
(No Transcript)
14
Source MCDDB (Beta Version)
15
(No Transcript)
16
Source MCDDB (Beta Version)
17
Source MCDDB (Beta Version)
18
Source MCDDB (Beta Version)
19
Master Production Schedule
Source MCDDB (Beta Version)
20
Resource Utilization
Source MCDDB (Beta Version)
21
Supply Chain Modeling and Optimization Using Soft
Computing Based Simulation
22
Introduction

• Supply chains involve the activity and
  interaction of many entities.
• Decision makers typically have imprecise goals.
• e.g. High service level
• Some system parameters may also be imprecise.
• e.g. Production capacity
• Discrete event simulation can help design and
  analyze supply chains.
• Many configurations and courses of action need to
  be investigated.
• Even experts have to spend a considerable amount
  of time searching for good alternatives.
• Soft computing guided simulation speeds up the
  process.

23
Scheme
Knowledge Extraction
Supply Chain Configuration
Input - Performance Data
Simulation
Goals met?
Yes
Stop
No
Fuzzy System / Relationship Identification
Activate Fuzzy Rules/Logic
Soft Computing Guided Simulation
SCBS (Alpha Version)
24
Knits
Retailer 1
Distribution Center 1
Distribution Center 2
Retailer 2
Wovens
Source SCBS (Alpha Version)
25
Knits
Retailer 1
Distribution Center 1
Distribution Center 2
Retailer 2
Wovens
26
Linguistic Terms

• Factory
• Production rate (low, medium, high)
• Finished inventory (small, medium, large)
• Utilization (low, medium, high)
• Distribution Center
• Inventory level (small, medium, large)
• Demand Point / Retailer
• Demand rate (low, medium, high)
• Service level (low, medium, high)
• Changes in inventory limits
• Large drop, Small drop, No change, Small
  increase, Large increase
• Changes in production rates
• Large reduction, Small reduction, No reduction,
  Small increase, Large increase

27
Source SCBS (Alpha Version)
28
Rule base to guide supply chain reconfiguration

• Example rule 1
• If
• Inventory level in the Distribution Center 1 is
  High
• and
• Inventory level in the Factory (Wovens) is
  Medium
• then
• Change in production rate in the Factory
  (Wovens) is Small Reduction.
• Example rule 2
• If
• Service level in Retailer 1 is Low
• and
• Inventory level in the Distribution Center 1 is
  Low
• then
• Change in production rate in the Factory (Knits)
  is Large Increase.

29
Goals

• The degree of fulfillment of the goals can be
  evaluated. e.g.
• Goal 1 High Service Level in Retailer 1.
• Goal 2 Low Inventory Level in Retailer 2.
• Goal 3 Medium Inventory Level in Factory
  (Knits).
• Each goal is met to a certain degree.
• A complicated a multi-criteria objective can be
  specified using AND, OR, NOT operators,
• e.g.
• High S.L. in Retailer 1 and Low Inventory Level
  in Retailer 2 and not Low Throughput in Retailer
  1 and Low Finished Inventory in Factory (Knits).

30
System / Relationship Identification
31

• In the previous scheme

Knowledge
Supply Chain
Extraction
Configuration
Input - Performance
Simulation
Data
Yes
Goals
met?
Stop
No
Fuzzy
Activate Fuzzy
System / Relationship
Rules/Logic
Identification
Soft Computing
Guided Simulation
Source SCBS (Alpha Version)
32
Methodologies for System Identification

• Conventional Mathematics
• Fuzzy Systems
• Neural Networks

33
Flow Chart
34
Neural Networks

• Idea
• to approximate the relationship between input and
  output data pairs.
• Steps
• train the neural network with existing data.
• predict performance using trained neural network.

35
Network Architecture
36
Test Case
Truck-Backup Problem
37
Results Comparison
38
Decision Surface Modeling
39
Decision Surface Modeling(Retail Model
Sourcing Simulator)

• Objective is to provide an interactive system
  that captures the essential features of the
  retail model in multidimensional, mathematical
  relationships between performance measures,
• e.g., service level, and key parameters, e.g.,
  reorder leadtimes.
• Provide retailer with a rapid, easy-to-use,
  visual tool to help understand and predict the
  impact on system performance of what-if
  scenarios such as
• What are the consequences of reducing initial
  season inventory?
• What are the consequences of poor forecast?
• What are the cost/benefits of reducing reorder
  lead times?

40
Source Sourcing Simulator (Version 2.0)
41
Neural Network Architecture

• A neural network consists of several layers of
  computational units called neurons and a set of
  data-connections which join neurons in one layer
  to those in another.
• The network takes inputs and produces outputs
  through the work of trained neurons.
• Neurons usually calculate their outputs as a
  sigmoid, or signal activation function of their
  inputs,

• Using some known results, i.e., input-output
  pairs for the system being modeled, a weight is
  assigned to each connection to determine how an
  activation that travels along it influences the
  receiving neuron.
• The process of repeatedly exposing the network to
  known results for proper weight assignment is
  called training.

42
Learning Curve
Source Sourcing Simulator (Version 2.0)
43
Offshore vs QR Sourcing Service Level Performance
Source Sourcing Simulator (Version 2.0)
44
Early Peak vs Late Peak Demand Gross Margin
Performance
4
4
Source Sourcing Simulator (Version 2.0)
45
Service Level -- Adjusted Gross MarginTradeoff
Adjusted Gross Margin
Service Level
Source Sourcing Simulator (Version 2.0)
46
Confidence Intervals for Estimated Decision
Surfaces

• A new approach based on jackknifing promises to
  yield reliable, realistic confidence and
  prediction bands on the estimated response
  surface.
• With k replications (runs) of n training patterns
  (design points), we combine k1 response surface
  estimates to obtain both point and confidence
  interval estimates of the average response
  EY(x) at each selected combination x x1, x2,
  ..., xm of the m decision variables.
• On the jth replication of all training patterns,
  common random numbers are used to sharpen the
  estimation of the ANN weights but as usual,
  different replications are mutually independent.

47
New Jackknife Procedure

• Let Y(x) denote an ANN estimate of the average
  simulation response at design point x when all k
  replications of the simulation are included in
  the training data.
• Let Y-j(x) denote the ANN estimate of the average
  simulation response when the jth replication of
  each training pattern is deleted.

48
New Jackknife Procedure

• The jth pseudovalue is ZjkY(x)
  (k1)Y-j(x) and from the sample mean Z and the
  sample standard deviation SZ of the pseudovalues,
  we compute the 100(1?) confidence interval for
  the expected response at design point x
  
  Z ? t1-?/2 k-1 SZ /?k.

49
(No Transcript)
50
Confidence Intervals for Estimated Decision
Surfaces

• A new approach based on jackknifing promises to
  yield reliable, realistic confidence and
  prediction bands on the estimated response
  surface.
• With k replications (runs) of n training patterns
  (design points), we combine k1 response surface
  estimates to obtain both point and confidence
  interval estimates of the average response
  EY(x) at each selected combination x x1, x2,
  ..., xm of the m decision variables.
• On the jth replication of all training patterns,
  common random numbers are used to sharpen the
  estimation of the ANN weights but as usual,
  different replications are mutually independent.

51
(No Transcript)
52
(No Transcript)
53
(No Transcript)
54
Fuzzy Inventory Replenishment System
55
Fuzzy Inventory Replenishment System
Centennial Apparel
56
Replenishment Policy

• Crisp (Q,R) System
• Q purchase amount
• R re-order point

Policy whenever inventory is below R, purchase
Q

• Fuzzy (Q,R) Control System
• Fuzzy Variables
• Membership Functions
• Fuzzy Rules
• Approximate Reasoning Method
• Defuzzification Method

Policy apply fuzzy rules to specify purchase
quantity
57
Fuzzy (Q,R) Inventory System
Linguistic Variables Inventory Low, Medium,
High Demand Low, Medium, High Purchase
Low, Medium, High
58
Fuzzy (Q,R) Inventory System
Membership Functions Trapezoid Numbers
Source Fuzzy (Q,R) (Beta Version)
59
Fuzzy (Q,R) Inventory System
Fuzzy Rules
Source Fuzzy (Q,R) (Beta Version)
60
Case Study
Source Fuzzy (Q,R) (Beta Version)
61
Case Study
Statistics from 50 Runs
Source Fuzzy (Q,R) (Beta Version)
62
Case Study
Modify Fuzzy Rules
Source Fuzzy (Q,R) (Beta Version)
63
Case Study
Performance Improved
Source Fuzzy (Q,R) (Beta Version)
64
More to come...http//www.ie.ncsu.edu/fangroup/