IEOR250: Introduction to Production and Logistics

Models

⃝2010
c Phil Kaminsky
ii
Acknowledgement

This material was initially typed from lecture notes taken in the Fall 2004 semester by Elise
Leduc, David Lewkowski, and Julien Clark. The 2010 revision is based on scribe notes taken
by students in the 2009 semester.

iii
iv
Contents

Acknowledgement iii

1 Introduction 1
1.1 What Is Logistics? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 What About Supply Chain Management? . . . . . . . . . . . . . . . 1
1.2 The Structure of Production/Distribution Networks . . . . . . . . . . . . . . 2
1.2.1 Serial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Converging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.3 Diverging (Arborescent) . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.4 Why Do We Have Intermediate Stages? . . . . . . . . . . . . . . . . . 3
1.3 Objectives and Constraints(customer value and costs) . . . . . . . . . . . . . 3
1.3.1 The Goal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.2 Customer Value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3.3 Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Inventory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.1 Is Inventory Necessary? . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4.2 Economic Order Quantity . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.1 Adding Uncertainty to the EOQ Model . . . . . . . . . . . . . . . . . 10
1.6 Uncertainty and Production Planning . . . . . . . . . . . . . . . . . . . . . . 11
1.6.1 Data: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6.2 Parameter Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6.3 Variable Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6.4 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6.6 New Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6.7 New Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6.8 Updated Model (Stochastic with Recourse) . . . . . . . . . . . . . . . 14
1.6.9 Updated Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6.10 First Model Actuals . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.6.11 Percent improvement from using updated model over original model . 15
1.7 More Production Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.7.1 EOQ Extension: Production Rate > Demand Rate . . . . . . . . . . 15

v
 1.7.2 Multiple Jobs, Each Takes Time, Shared Resource . . . . . . . . . . . 16
1.8 Physical Flows and Transport . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.9 Information and Decisions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.10 Time Horizons and Hierarchical Levels . . . . . . . . . . . . . . . . . . . . . 17
1.10.1 Strategic planning (3-5 years ) : . . . . . . . . . . . . . . . . . . . . . 17
1.10.2 Tactical planning (6months-1 year ): . . . . . . . . . . . . . . . . . . 17
1.10.3 Operational control : . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.11 Decision Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.12 Quantitative Models and Methods . . . . . . . . . . . . . . . . . . . . . . . . 18
1.12.1 Model: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.12.2 Mathematical Model: . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.12.3 For Prescriptive Models: . . . . . . . . . . . . . . . . . . . . . . . . . 19

vi
Chapter 1

Introduction

1.1 What Is Logistics?

The Council of Supply Chain Management Professionals defines Business Logistics as follows:

”Business Logistics is the process of implementing and controlling the efficient,

cost effective forward and reverse flows of goods, services, and information be-
tween the point of origin and the point of consumption to meet customer require-
ments.”

1.1.1 What About Supply Chain Management?

Similar (maybe the same as) logistics, sometimes broadened to include more of a focus on:
• Total system cost

• Additional aspects including:

– integration supplier relationship management

– contract design
– customer relationship management
– information technology/ ERP
– marketing
– finance
– product design
– risk and globalization
∗ unknown-unknown
– revenue and yield management
– environmental issues

1
 ∗ reverse logistics
∗ carbon footprint

1.2 The Structure of Production/Distribution Networks

1.2.1 Serial
This is the most basic network that shows different stages of supply chain, such as produc-
tion, assembly, packaging, storage and consumption.

Figure 1.1: Serial Network

1.2.2 Converging
This is a network that models assembly.

Figure 1.2: Converging Network

Third type of distribution networks is the diverging (arborescent) network as shown in

figure 1.3.

1.2.3 Diverging (Arborescent)

This is a pure distribution network.

2
 Figure 1.3: Diverging Network

1.2.4 Why Do We Have Intermediate Stages?

• Economies of Scale

• Pooling

• Cross-Docking

1.3 Objectives and Constraints(customer value and costs)

1.3.1 The Goal
Appropriate (good, suitable) service at minimal cost

1.3.2 Customer Value

• quality

• availability

– make-to-stock (MTS) items: fraction in stock

– make-to-order (MTO) items: cost/lead time trade-off

• assortment

• flexibility

1.3.3 Cost
All models trade off costs and benefits. So what are the costs?

3
Cost Classification
• linear vs. nonlinear
∑N
– Linear: f (x) = cx (marginal cost = c); f (x) = l=1 cl xl
√
– Nonlinear: f (x) = x; f (x1 , x2 ) = x1 x2

• fixed cost vs. variable cost

{
F + Cx ifx > 0
– example: c(x) =
0 otherwise

• semi-variable cost: (See picture – example is shipping cost)

Figure 1.4: Semi-variable Cost

What Are the Key Costs?

• Facility-related costs: Building costs which are fixed costs and depend on capacity,
location and type of facility.

• Transportation cost: They are complicated costs, but linear functions approximate
them reasonably well.

• Inventory-related costs:

– Ordering cost: The ordered volume affects ordering price.

– Inventory holding cost: It depends on the opportunity cost, storage cost, insurance
cost, material handling cost and expiration costs. This cost is linear in inventory
level.
– Shortage cost: It is mainly composed of two parts: Lost sales and loss of goodwill.
Loss of goodwill is very hard to assess.

4
1.4 Inventory
1.4.1 Is Inventory Necessary?
• Too much inventory may be bad for reasons including storage costs, the tying up
of resources in the inventory, and the possibility of expiration or obsolescence of the
materials stored as inventory.

• Some inventory may be required in order to meet target service levels.

• The goal is to find the minimum inventory you need, and to figure out how to reorganize
operations so that you need less.

Characteristics of Systems That Don’t Need Inventory

• no lead time

• no economies of scale in production or transportation

• no uncertainty

• no interaction of inventory with demand. (Alternatively, inventory may help drive

sales. For example, full shelves at a store may help attract customers.)

Modeling
In this course, we will put much emphasis on quantitative inventory models, starting with
EOQ. In addition to thinking about how much inventory to maintain and when to place
orders, we should think about where in the supply chain the inventory is held (near the
front, near the back, ...)

1.4.2 Economic Order Quantity

The Economic Order Quantity (EOQ) model is a simple starting point for modeling inventory
and resupply. The model is widely used (perhaps too indiscriminately).
This model was developed in 1913 by F.W. Harris:
Harris, F.W. (1913) ”How Many Parts To Make At Once” Factory, The Magazine of
Management, 10(2), 135-136, 152.

The Model
• Assume constant demand, at d units per unit time.

• Because the demand is constant, it is reasonable to order the same quantity, Q, each
time we place an order.

5
 • Initially, assume no lead time. (orders are delivered immediately after being placed)
• Assume without loss of generality (WLOG) that at time t = 0 the Inventory = 0, so
an order is placed for quantity Q.
• Without a lead time for the orders, a time series for the inventory looks like figure 1:
Q
Inventory

0 Q/d 2Q/d 3Q/d

time

Figure 1.5: Inventory vs Time with no lead time

• Stipulate that we must meet all demand.

With this setup, there are two costs associated with an inventory policy:
k + cx = ordering costs
hx = holding costs
where,
x = quantity ordered
k = fixed ordering cost
c = cost per item
h = holding cost per item per unit time

What If There Is A Lead Time?

Let L be the lead time – the time between placing an order and receipt of the order.
• Assume L < Q
d

• In order to receive the order just as inventory is depleted, place the order when time
t = Qd − L

6
 • Equivalently, place the order when the inventory level = dL.

• With this policy, orders will be place when there is just enough inventory to meet
demand during the lead time.

What Is the Cost Of a Cycle?

The cost of a cycle between orders of size Q is as follows:

cost = ordering costs + holding costs

= k + cQ + h(area under curve [inventory vs time])
= k + cQ + h(average inventory ∗ cycle duration)
Q
= k + cQ + h 12 Q
d

What Is The Cost Per Unit Time?

Divide the cost per cycle by the time per cycle to get the cost per unit time, T C(Q):

[ ]
1 hQ
T C(Q) = Q k + cQ + Q
d
2d
kd hQ
T C(Q) = + cd +
Q 2

We can now minimize T C(Q) with respect to Q:

kd h
T C ′ (Q) = − +
Q2 2
2kd
T C ′′ (Q) = >0
Q3

so setting the first derivative to zero and solving for Q minimizes T C(Q):

kd h =
− + set 0
Q2 2
√
2kd
Q =
h
Note,
√
∗
• Q = 2kd
h
does not depend on the cost per item, c.

7
 • In the long run, the amount we order is equal to demand, regardless of the cost per
item, because we must meet the demand – while holding costs do not depend on item
costs.
• Trends:
– order size Q increases as fixed cost of placing an order, k, increases
– order size Q increases as demand, d, increases
– order size Q increases as holding cost, h, decreases
– the optimal inventory policy balances holding costs and ordering costs
• Total cost per unit time with the optimal order size Q∗ :
kd hQ∗
T C(Q∗ ) = + cd +
Q∗ 2
√ √
h h 2kd
= kd + cd +
√ 2kd √2 h
kdh kdh
= + cd +
√ 2 2
= 2kdh + cd

Extensions to the EOQ

The Economic Order Quantity model may be extended in many ways:
• More than one item
• Finite time horizon
• Changes in demand
• Random demand
• Random demand with the option of back orders – where we can meet present demand
by paying a penalty and delivering items to customers in the future. In this case, a
policy will be set to balance holding costs versus the penalty on back orders. Policies
with back orders can become complicated when we consider interactions between back
orders, the customers’ good will, and marketing.

Robustness
In real life, the assumptions of our models often do not hold, and we have to estimate the
values of model parameters. So, we should consider how robust our models are to failed
assumptions or misspecified parameter values.
For example, we may consider how robust the total cost function in the EOQ model,
T C(Q), is to selection of an order quantity Q′ that is not the optimal quantity Q∗ .

8
 Let Q′ = bQ∗ .
Then,
√ √
′ kd h hb 2kd
T C(Q ) = + cd +
b√ 2kd 2√ h
1 kdh h 2kd
= + cd + b
b 2 √ 2 h
( )
1 kdh
= +b + cd
b 2
We can now calculate a relative error to compare this total cost to the optimal total cost:
(1 ) √1 √
′
T C(Q ) − T C(Q ) ∗
b
− b 2
kdh − 2kdh
= √
T C(Q∗ ) 2kdh + cd
(1 ) √1 √
b
− b 2
kdh − 2kdh
≤ √
( ) 2kdh
1
= 12 +b −1
b

Here we see that if b = 1.5, so that Q′ is 50 percent too large, the relative error in the total
0.25
0.20
{ TC(Q’) − TC(Q*) } / { TC(Q*) }

0.15
0.10
0.05
0.00

0.6 0.8 1.0 1.2 1.4

Figure 1.6: Relative error in T C(Q′ ) as a function of b for Q′ = bQ∗

cost per unit time, T C(Q) is only 8.3 percent. So, in this case, the inventory policy performs
OK if the order quantity Q is anywhere near he optimal value.

9
1.5 Uncertainty
There are many uncertainties in logistics:

• in supply

• in demand

• in prices

• in exchange rates

• in yields (particularly in biotech and semiconductor industries, where batches have

variable yield, and occasionally entire batches must be discarded)

• in competitor performance

• in future capacity

• in many other factors

In this course, the type of uncertainty that we will deal with are ”known unknowns” –
cases where we can place probability distributions on uncertain features of our models.

1.5.1 Adding Uncertainty to the EOQ Model

We could, for example, model uncertain demand. This will affect the size of the orders we
place and the timing of the orders. For now assume that we still order a quantity Q, but in
other models we will jointly determine the order size Q and the reorder point R.
In the case of constant demand with lead time, an order is placed when inventories drop
to the level dL = R (reorder point).
Now, let demand during lead time be randomly distributed, with a Normal distribution
having mean µL and standard deviation σL :

DL ∼ N (µL , σL )

With this model, in order to always be able to meet customer orders, we would need
an infinite inventory. The probabilities of very large spikes in demand during lead times are
low, so instead of guaranteeing that we will always be able to meet demand, we will set an
inventory policy that will give an appropriate service level.
For example, we can set a reorder point R such that the probability of running out of
stock during the lead time is α.
Equivalently, the probability of the demand during lead time, DL , being less than or
equal to R is 1 − α:

Pr (DL ≤ R) = 1 − α

10
 Using the Normal model for demand during lead time, we can solve for the reorder point:

R = µL + Z(1−α) σL
Here, the second term contributing to the reorder point, Z(1−α) σL , represents how much
extra inventory we need to have on hand when we place a new order because of variability
in demand during lead time.
The increase in total costs are due to increases in holding costs in the variable demand
case.

How to lower costs when faced with variable demand

Now, instead of considering a variable demand over the entire lead time, we can consider the
(variable) demand per day:

d ∼ N (µ, σ)
If the lead time is L days and we assume that the demands during each day are indepen-
dent, the total demand over the lead time will be
( √ )
DL ∼ N Lµ, σ L
The greater the standard deviation of the demand during lead time, the greater the
amount of inventory we need to have on hand when a new order is placed, and the greater
the storage costs. But this variation in the demand during lead time may be decreased by
reducing the lead time – decreasing L.
The total cost can also be decreased by reducing the customer service level or decreasing
the variance in demand.

1.6 Uncertainty and Production Planning

In the previous model, we set the policy and it runs. Sometimes, we need to adjust as
uncertainty is resolved...
ATO Model (Assemble to Order)
- Assemble products to meet demand
- Make components before we see demand

11
1.6.1 Data:

Table 1.1: Bill of Materials

Product C1 C2 C3 C4 C5
A1 1 1 1 0 0
A2 1 1 0 1 0
A3 1 1 0 0 1

Table 1.2: Machine Times and Component Costs

Component M1 M2 M3 Cost
C1 1 2 1 20
C2 1 2 2 30
C3 2 2 0 10
C4 1 2 0 10
C5 3 2 0 10
Avail. 800 700 600

Table 1.3: Equal Prob. Demand Scenarios and Product Prices

Product S1 S2 S3 Avg Price
A1 100 50 120 90 80
A2 50 25 60 45 70
A3 100 110 60 90 90

1.6.2 Parameter Definitions

Gij - Number of component i for product j

Tim - Time needed for one unit of component i on machine m

Ci - Cost per unit of component i

dj - Average demand for product j

pj - Unit selling price for product j

Lm - Total capacity of machine m

12
1.6.3 Variable Definitions

Xi - Amount of component i to be made

Yj - Amount of product j to be assembled and sold

1.6.4 Model

∑
3 ∑
5
Max pj Yj − c i Xi j = 1..3, i = 1..5
j=1 i=1

∑
5
Subject to: Tim Xi ≤ Lm m = 1..3
i=1
Yj ≤ d¯j j = 1..3
∑
3
Gij Yj ≤ Xi i = 1..5
j=1

Y j , Xi ≥ 0 j = 1..3, i = 1..5

1.6.5 Results

X1 = 116.67, X2 = 116.67, X3 = 26.67, X4 = 0, X5 = 90

Y1 = 26.67, Y2 = 0, Y3 = 90

Objective = 3, 233.33

The above model does not take into account the fact that we have to make components
before observing demand, but we don’t have to assemble finished products until after we
observe demand.

1.6.6 New Parameters

dsj - Demand for product j under scenario s

πs - probability of event s occurring, in this case πs = 1/3 for s = 1..3

13
1.6.7 New Variable

Yjs - Amount of product j under scenario s

1.6.8 Updated Model (Stochastic with Recourse)

∑
3 ∑
3 ∑
5
Max πs pj Yjs − c i Xi j = 1..3, i = 1..5, s = 1..3
s=1 j=1 i=1

∑
5
Subject to: Tim Xi ≤ Lm m = 1..3
i=1
Yjs ≤ dsj j = 1..3, s = 1..3
∑ 3
Gij Yjs ≤ Xi i = 1..5, s = 1..3
j=1

Yjs , Xi ≥0 j = 1..3, i = 1..5, s = 1..3

1.6.9 Updated Results

X1 = 115.71, X2 = 115.71, X3 = 52.86, X4 = 2.86, X5 = 62.86

Table 1.4: Updated Results

S1 S2 S3
Y1 52.86 50 52.86
Y2 0 2.86 2.86
Y3 62.86 62.86 60

Objective = 2, 885.71

1.6.10 First Model Actuals

We would never actually produce the quantities of final products given by the first model.
If we used the component levels given by the first model, here is the final product quantities
we would produce:

Expected profit of these scenarios = 2, 333.33

14
 Table 1.5: Updated Results
S1 S2 S3
Y1 26.67 26.67 26.67
Y2 0 0 0
Y3 90 90 60

1.6.11 Percent improvement from using updated model over orig-

inal model

(2, 885.71 − 2, 333.33)/2, 333.33 = 23.67%

1.7 More Production Models

Need to account for the fact that production takes time, products are often made in sequence
on a shared resource.

1.7.1 EOQ Extension: Production Rate > Demand Rate

Assumptions:

• Same as EOQ.

• After order, receive at rate P .

15
 Observe:

Q(P −D)
T = Q
D
, Tp = Q
P
, I = Tp (P − D) = P

Costs calculation:

Cycle cost for a cycle of length T : K + CQ + hT2p I + hT2D I = K + CQ + hT I

2
Cycle cost per unit time : KD
Q
+ CD + hI2
(Q = T D)

−Q)
T C(Q) = KD Q
+ CD + hQ(P 2P
T C ′ (Q)= − KD + h(P2P−D) = 0
Q2 √ √
⇒ Q∗ = 2KD
h
P
P −D
T C ′′ (Q)= 2KD
Q3
>0

1.7.2 Multiple Jobs, Each Takes Time, Shared Resource

∑
Simple Goal Given job i with processing time pi , weight wi , minimize j wj Cj
Assumption:
rj = 0 for all j (release date = 0 )

WSPT rule (Weightened shortest processing time) :

w
Sequence jobs in decreasing order of pjj .

Theorem :
WSPT rule is optimal for minimizing sums of weighted completion times.

Proof :
Suppose by contradiction that in optimal schedule S, these are two jobs such that k
w w
follows j and pjj < wpkk (because k follows j, we should have pjj > wpkk ).
Interchange these two jobs. No other jobs change so, if job j starts at t :

• current objective is : (t + pj )wj + (t + pj + pk )wk

• new objective is : (t + pk )wk + (t + pj + pk )wj

Therefore the change in objective is: pj wk − pk wj > 0 since pk wj < pj wk.

This contradicts our initial assumption of optimality.
So the objective decreases.

16
1.8 Physical Flows and Transport
• Once network exists, need to transport goods

• Key decison: mode selection

• Key decision: point-to-point vs. multiple destinations

• Key Model: The Vehicle Routing Problem

– Key subproblem: The Traveling Salesperson Problem

1.9 Information and Decisions

• Information is key to logistics

• Centralized decision-makers can optimize

• Sometimes this is impossible – why?

1.10 Time Horizons and Hierarchical Levels

All decisions connected, but global optimization is often impossible. It is often useful to
think about:

1.10.1 Strategic planning (3-5 years ) :

• Product line selection

• Facilities design (long lasting effect, long planning horizon)

• Vehicle purchases

• Supplier qualification...

1.10.2 Tactical planning (6months-1 year ):

• Allocating resources to satisfy demand

• Production rates, work force size, transportation contracts

• Inventory policies...

17
1.10.3 Operational control :
• Ensuring specific tasks are carried out efficiently

• Daily policy parameters, order decisions, ...

This view suggests a hierarchical approach to decision making.

1.11 Decision Approaches

Push vs. Pull
• Push: Forecast-driven decisions

• Pull: Demand-driven decisions.

Make-to-order vs. Make-to-stock

1.12 Quantitative Models and Methods

We will focus on using mathematical models to address these issues.

1.12.1 Model:
A representation of a system that contains sufficient characteristics to get useful information
out.
Note:

• What is useful information ?

• “All models are wrong, some are useful”

1.12.2 Mathematical Model:

A symbolic or abstract representation of a system.
→ Descriptive models : predict system performance
→ Normative/Prescriptive models : optimize set of decision variables

18
1.12.3 For Prescriptive Models:
• Optimal solution: Best possible (found by optimal algorithm).

• Heuristic algorithm:

– Finds good feasible solution.

– Can we say how good a heuristic solution is, without knowing the optimal solu-
tion?
∗ We can use bounds.
∗ Worst case analysis
∗ Average case analysis

19