Chapter 8: Multiechelon Inventories

¨ Characteristics
¨ Lot Sizing with Multiple Echelons
¨ Power-of-Two Policies (PO2)
 1. Characteristics

 Items with Independent demand

 Items with dependent demand

 3

Example 1: A Series System

Customer demand

External supply

Item 1 Item 2 Item N

 Example 2: An Assembly System

Customer demand
External supply

2 6 7

3 5
 Example 3: A Disassembly System

3 5

Customer demand
1 2
6
External
supply
4

7
Example 4: A Distribution System

Customer demand
3

External supply
4

5
 Other Examples

 Disassembly/assembly systems

 Assembly/distribution systems

 Distribution systems with transshipments

 Distribution systems with multiple supply

sources
2. Lot Sizing with
Multiple Echelons
 Example
Product 1 Product 2

Intermediate G
Feed A
Reactor Reactor Reactor

Intermediate E

Feed B Intermediate F
Intermediate H

Reactor Reactor

Feed C

Feed D
 The Item-Task Network Representation
End Product 1 End Product 2

Intermediate G

Feed A Task Task

Task 2 Task
Task 5
1 2 5
Intermediate E

Feed B
Intermediate F

Task
Intermediate H
Task 4
3
10
Feed C
Feed D
 Items & Tasks
 An item can be a component purchased from an outside
supplier or produced internally.
 An item can be a raw material (e.g., a component), a semi-
finished (e.g., sub-assembly) or a finished product.
 A task can consume and produce multiple items
(components/intermediates/products).
 An item can be consumed by more than one task;
similarly, an item can be produced by more than one task.
 A finished product can be the result of several tasks done
in series or in parallel.
 Example

7
5
6
Task
4 3
1 2 4 5
3 8
Task
1
1
8 7
2 5 1
5
Task
4
Task 9
2 1
6
5
3
 System Description
 t: a period (e.g., day, week, month); t = 1, … ,T, where T
represents the planning horizon
 Drt: demand for item r in period t (number of units), r=1,…,
R where R is the number of items
 rir: number of units of item r needed to carry out task i, i=1,
…, N where N is the number of tasks
 sir: number of units of item r produced by task i
 Problem Statement

 Given a demand profile over a set of T periods for

each item (demand can be for either finished or
semi-finished products), determine the quantity of
each item to produce in each period in order to
minimize the production, inventory and setup
costs, while meeting demand and without
exceeding production capacity.
 Formulations

 Big bucket formulation (the production

planning problem)

 Small bucket formulation (the production

scheduling problem)
 A Big-Bucket Formulation
 Tasks initiated in a period are completed during the same
period
 The same task can be carried out multiple times during a
period
 Items produced in a period can be used to satisfy demand
during that period
 Demand in each period must be satisfied in that period; no
backorders allowed
 There are no capacity limits (no limits on the number of times
a task can be carried out in a given period)
 Notation

Parameters
 cit: variable cost of carrying out task i in period t, i=1,…,
N where N is the number of tasks
 Ait: fixed cost of carrying out task i in period t (incurred
at most once during each period)
 hrt: cost of holding one unit of item r in inventory from
period t to period t+1
 Notation (Continued…)

Decision variables
 Irt: inventory level of item r at the end of period t

 Qit: the number of times task i is carried out in period t

 Yit = 1 if task i is initiated one or more times during

period i and Yit = 0 otherwise

 Notation (Continued…)


N
 Amount of item r produced in period t = i 1
 irQit


N
 Amount of item r consumed in period t =
i 1
 irQit
 Formulation

Minimize z   t 1  i 1  cit Qit  Ait Yit    t 1  r 1 hrt I rt

T N T R

subject to
I rt  I r ,t 1   i 1 ir Qit   i 1 ir Qit  Drt ,  t , r
N N

Qit  MYit i, t , r

Yit {0 ,1} i, t
Qit , I rt  0 i, t , r
 Formulation

z   t 1  i 1  cit Qit  Ait Yit    t 1  r 1 hrt I rt

T N T R
Minimize
subject to
I rt  I r ,t 1   i 1 ir Qit   i 1 ir Qit  Drt ,  t , r
N N

Qit  MYit i, t , r

Yit  {0 ,1} i, t
Qit , I rt  0 i, t , r

M is a large number
 A Formulation with Capacity
Constraints and Multiple Machines

 A machine may correspond to a single processor, an

assembly workstation, or a production line, among
others
 It is possible for a task to be carried out on one or more
machines
 A machine could possibly carry out more than one task
 A machine has a finite capacity and carrying out any
task consumes some of this capacity
 Notation

 Utm: capacity of machine m in period t; m = 1, … , M,

where M is the number of machines
 aim: units of capacity of machine m needed to carry out
task i
 ci,t,m: variable cost of carrying out task i on machine m in
period t
 Ai,t,m: fixed cost of carrying out task i on machine m in
period t
Note: A task that cannot be carried out on a
machine is assigned a very large production
cost.
 Notation (Continued…)

 Qi,t,m: the number of times task i is carried out in

period t on machine m

 Yi,t,m = 1 if task i is initiated one or more times during

period i on machine m and Yi,t,m = 0 otherwise

 Formulation

 t 1  i 1  m1  ci ,t ,mQi ,t ,m  Ai ,t ,mYi ,t ,m   t 1  r 1 hrt I rt

T N M T R
Minimize z

subject to
I rt  I r ,t 1   i 1  m 1 irQi ,t ,m   i 1  m 1  irQi ,t ,m  Drt ,  r, t
N M N M


N
i 1
 imQi ,t ,m  U tm  t, m
U tmYi ,t ,m
Qi ,t ,m   i, t , m
 im
Yi ,t ,m  {0,1}  i, t , m
Qi ,t ,m , I rt  0  i , r , m, t
A Formulation with Setup Times

 A setup time si,m is incurred if task i is carried out on

machine m, one or more times in any given period.

 The capacity constraint is modified as follows


N
i 1
 imQi ,t ,m  simYi ,t ,m  U tm  t, m
 A Small-Bucket Formulation

 Time periods are chosen to be small enough so that only one

task on a particular machine can be either initiated or
completed
 The processing time of each task consists of one or more
periods
 A setup cost is incurred when the task initiated on a machine
is different from the task that was just completed on that
machine.
 Notation

 ai,m: processing time (in number of time periods) of task i on

machine m

 Qi,t,m = 1 if task i is initiated on machine m at time t, and Qi,t.m

= 0 otherwise

 Zi,t,m = 1 if machine m at time t is set up for task i

 Formulation
 m1  t 1  i 1  ci ,t ,mQi ,t ,m  Ai ,t.mYi ,t ,m    t 1  r 1 hrt I rt
M T N T R
Minimize z
subject to
I rt  I r ,t 1   m 1  i 1 irQi .t i ,m ,m   m 1  i 1  irQi ,t ,m  Drt ,
M N M N
r, t

 Q
i , m
u 1 i ,t  u 1,m  Z i ,t , m i, t , m

 Z
N
i 1 i ,t , m 1 t , m
Yi ,t ,m  Z i ,t ,m  Z i ,t 1,m i , t , m
Qi ,t ,m & Z i ,t ,m  {0,1} i , t , m
I rt  0 r, t
0  Yi ,t ,m  1 i , t , m
 Solution Methods

 Small to medium problems can be solved exactly (to

optimality)
 Large problems may not solve within a reasonable
amount of time (the problem belongs to a class of
combinatorial optimization problems called NP-hard)
 Large problems can be solved approximately using a
heuristic approach
 Example Heuristics

 Decompose the problem into a series of sub-problems

with smaller planning horizons
 Decompose the problem into a series of subproblems,
each pertaining to a single item
 Decompose the problem into one big bucket problem
(the production planning problem)
 Solve the problem by relaxing one or more sets of
constraints
 Solve the problem on a rolling horizon basis
 3. Power-of-two Policies
¨ To deal with multiple location

¨ Some changes in notations:

- Demand rate is 𝜆 instead of D

- Holding cost is h (as usual)

- Ordering cost is K instead of A

- Cycle time or reorder interval is T = Q/𝜆

- T will be the decision variable instead of Q

- Total cost is Z(T); g=0.5*h*𝜆

 Power-of-two Policies
¨ T policy is sometimes impractical to implement (e.g., T* is
1.09 weeks) ⟹ using base planning period TL
¨ Reorder interval is an integer multiple of TL

¨ If we constraint n to be the power of two

 Why power of two (PO2)?
¨ The worst-case cost of using PO2 policy is lower than the worst-case cost
of using PO3, PO4, etc.
¨ PO2 policy’s worst-case cost is closest to the optimal cost.

Z
(1)
Z

(2)
Z A Power of Two time interval is
guaranteed to be within 6% of
(3) the costs with the optimal time
Z interval.
 Proof
36
¨  (1) :
¨ (2) :Reorder interval must be of the form

for some k ∈ { 0, 1, 2, …}
¨ Optimal order interval is
 Optimal Power-of-Two T
37

¨ Let f (T) = EOQ cost as function of T:

¨ f is convex
¨ Therefore, optimal k is smallest k such that

…
 Optimal Power-of-Two T
38

¨ 

¨ Optimal order interval is T = TB2k, where k is smallest

integer such that

¨ (3): from (2) we have:

From (1) :
 Why power of two (PO2)?
¨ Easy to divide the production of items into two stages
¨ the week of the production was determined
¨ the exact time of the production (day, shift, etc.) was
determined
¨ Make production scheduling easier:

A, B, C produced
every 1, 3, 5 weeks
respectively

A, B, C produced
every 1, 2, 4 weeks
respectively
 3.1. PO2 for Single-Stage System
¨ TL is known, finding T* is equivalent to finding 𝓁*

subject to:

optimum condition:

optimal
reorder
interval
 Optimal Reorder Interval
for PO2

is optimal reorder interval from EOQ

¨ Example:
¨ TL is 1 weeks and TE is 0.196 years
 3.2. PO2 for Serial System

¨ This is a multi-stage extension of EOQ

¨ Each stage except stage n has exactly one predecessor stage

¨ Each stage except stage 1 has exactly one successor stage.

¨ Our goal is to determine how to determine the reorder intervals

for each of the n stages.
 Assumptions and Notations
¨ A serial system can be represented as a directed graph
¨ Stages form the nodes and the flow of the product is along
the arcs of this graph
¨ N(G) and A(G) denote the node set and the arc set,
respectively

and
 Assumptions and Notations
¨ 𝜆i represents the demand rate of stage i, measured in units per year
¨ Stage i can use several components produced at stage i+1
¨ 𝜆i can be different for each stage
¨ For simplicity, assuming that for 𝜆i = 𝜆 for all stage

reorder interval for each stage

¨ Only consider policies that are both nested (i.e. if stage i place an
order, all stage from 1 to i-1 also place order at the same time) and
stationary (reorder interval does not change over time)
 Costs in Serial Systems
¨ Fixed ordering cost Ki of stage i
¨ hi is the annual (on-hand) holding cost rate - quite hard to
used for complex network
¨ New approach: average holding cost at stage i = the average
echelon holding cost * echelon stock

¨average echelon holding cost is a measure of the

value added only due to the activity at stage i
 Costs in Serial Systems

the total value added up through stage i (of

which hi is the measure) is equal to the sum
of the incremental values added in stages n
through i (of which is the measure)
 Echelon Stock
¨ Echelon stock for stage i to be the sum of the on-hand
inventories at stages 1 through i
¨ Example: if stage 1 has 2 units, stage 2 has 3 units, and
stage 3 has 1 unit, then the echelon stocks for stage 3 and
stage 2 are equal to 6 units and 5 units, respectively. The
echelon stock for stage 1 is equal to its on-hand stock, 2
units.
 Echelon Stock
on-hand inventory echelon stock
Echelon Stock (cont.)
¨ average on-hand inventory
may be more difficult to
calculate as compared to
echelon inventory
¨ The two approach leads to
the same holding costs
 Echelon Stock (cont.)
assuming:

average annual inventory holding cost at stage 2 is:

average annual inventory holding cost for two stages:

 Echelon Stock (cont.)
annual inventory holding cost
at stage 2:

annual inventory holding cost

at stage 1:

average annual inventory holding cost for two stages:

 Nested Policy
¨ Nested policies are optimal for n-stage serial systems.
¨ Examples: two stage system
A Mathematical Model for Serial
Systems

subject to:
Relaxed Problem

easier to solve

RP ⇒ P
Subgraph

G is a subgraph of graph G′.

Also, N(G) = {1,2,3} and
A(G) = {(3, 2), (2, 1)}.
 Subgraph (cont.)

G1 and G2 are ordered by precedence if there does not exist an arc

(j,i) ∈ A(G′) such that j ∈ N(G1) and i ∈ N(G2)
 Ordered Partition
Definition of ordered partition G1, …, GM

Serial system can be easily represented as an order partition

A directed cut is just like any other ordered partition with one major
distinguishing feature: the ordered partition has only two subgraphs.
 Relaxed Problem (cont.)

subject to:

assuming that the reorder interval in Gk is the same = T(k)

subject to:
 Solution for RPk
it is similar to EOQ ⇒ take derivative, set it equal to zero:

or

where:
 Optimal conditions for RP

Theorem:
String Algorithm
 Example
The production of an item requires it to go through five processes
sequentially. The fixed cost of production at each stage is obtained by
multiplying the setup time with the labor cost per unit time. Annual
demand is 500.
fixed cost

annual holding cost

 Example (cont.)

Echelon holding cost

Example (cont.)
Example (cont.)
Example (cont.)
Example (cont.)
Example (cont.)
 Solving PO2 for
Serial System

subject to:
 Solution of
PO2 for Serial System
similar to single stage system, optimal value for ℓ is the smallest
non-negative integer such that

Solve the PO2 for serial system using 3-step procedure

.
 Example (cont.)
¨ Using the 5-step process example
¨ Assuming that the base planning period is 1 week
¨ For subgraph G1:

¨ For subgraph G2:

all stages will follow the same plan in the power-of-two solution