INDU 342: Logistics Network Models

Course Introduction

Claudio Contardo

Mechanical, Industrial and Aerospace Engineering

Concordia University

Lecture 1
About the instructor

Associate Professor at the MIAE Dept since August 2022

CPLEX Developer between Jan 2021 and July 2022
Associate Professor at UQAM between 2012 and 2020
PhD in CS from the U of Montreal, 2011
Engineering degree in applied mathematics, U of Chile, 2005

Office hours
Thursdays 9:00-11:00 EV 4.171
Logistics

Definition of logistics (Intro to Logistics Mgmt)

Logistics is the discipline that studies, in an organization (such as a
private company, a public administration, a non-profit association, a
military corps), the management and implementation of the operations
concerning the flow of tangible goods (materials, food and medical
supplies, refuse, equipment, weapons, etc.) from their sources (suppliers,
mines, crop fields, etc.) to their points of utilization or consumption or
disposal (retailers, landfills, army units, etc.) to meet the objectives of
the organization. To this end, logistics requires the collection,
integration, and processing of data from several sources in order to plan,
organize, and control activities such as material handling, production,
packaging, warehousing, and distribution
Some types of logistics systems I

Defense logistics is at the origin of the field of logistics. Babylonians

more than 4,000 years ago studied the deployment of
materials and infantry at times of war. WWII witnessed
the integration of logistics planning and computer science
Industrial logistics in manufacturing, where the concept of logistics
networks flourished. Concepts like suppliers, carriers,
logistics providers, wholesalers, retailers were born. In
here, we aim at planning the flows of materials and
resources from the acquisition until the delivery to the
final customers
Service logistics deal with the planning of activities (and of the resources
associated) to deliver services rather than tangible goods.
Postal services, technical services, urban solid waste
collection, are remarkable examples
Some types of logistics systems II

Integrated logistics and alliances deal with the planning of multiple

interconnected logistics problems at once. For instance,
integrating marketing with assortment planning and
production. When the integration involves multiple
organizations, we refer to it as a logistics alliance. The
airline industry is a remarkable example of Alliances.
Logistics systems

Logistics system
A logistics system is a set of interacting infrastructures, equipment, and
human resources whose objective is, as a whole, the execution of all the
functional activities determining the flow of materials among a number of
facilities. Facilities may be plants, warehouses, landfills, sorting centres,
air, and ground hubs where either production or assembly, disposal,
consolidation, storage, packaging, distribution, etc. is carried out.
Logistics systems

Figure: Example of a logistics system

Logistics systems

Figure: Network representation of a logistics system

Supply chains

Supply chain management

Supply Chain Management (SCM) is concerned with the coordination
and management of the supply chains of an organization
Taxonomy of supply chains I

Make-to-stock (MTS) When the demand of single products can be

predicted accurately, all the activities of procurement,
manufacturing, assembly, and distribution can be planned
in advance, based on forecasts of finished product demand
Assembly-to-order (ATO) When finished products come in a very large
number of variants and only the aggregate demand can be
predicted accurately, it is reasonable to produce common
components in advance (based on aggregate forecasts)
and then assemble the products when orders arrive
Make-to-order (MTO) When the aggregate demand of an entire range of
products is hard to be predicted, all the activities of
manufacturing, assembly and distribution (and sometimes
procurement) should be triggered by customers’ orders
Engineering-to-order (ETO) When some products are so unique that
even the design of the product is done on the basis of the
customers’ specifications
Taxonomy of supply chains

Material decoupling point

The material decoupling point (MDP) is the separation between the
activities based on planning (push subsystem) and the activities triggered
by orders (pull subsystem), which is often a significant stock-holding
pointbe. Products are pushed to the MDP and pulled from it
Taxonomy of supply chains

Figure: Taxonomy of supply chains

The bullwhip effect

The bullwhip effect

When forecasts for the demands of products are highly uncertain and
lead times important, we may observe stocks oscillating between
stockouts (due to underproduction) and large surplus (due to
overproduction) as a result, in the same season
The bullwhip effect is happening as we speak

Figure: Recent example of actions potentially leading to bullwhip effect

Case studies

Apple’s SC for iPhone

Adidas AG
Galbani
Pfizer
Amazon
FedEx
A.P. Moller-Maersk
Canadian-Pacific Railway
Trends in logistics

The last years have witnessed the raise of new trends in logistics
Reverse and sustainable logistics

Figure: Exemple of reverse logistics system

Trends in logistics

E-commerce and omnichannel logistics

City-logistics
Logistics objectives and KPIs I
We want to design logistics systems that are efficient
However, the meaning of being efficient is ambiguous. It may mean
different things to different people
Capital-related KPIs: Measure how well cash flow is managed
Cash-to-cash time (C2C): time from paying suppliers to receiving
cash from customers
Gross margin ROI (GMROI). gross profit / avg inventory investment
Inventory days of supply (IDOS): how many days until running out of
stock of a product
Inventory turnover (IT): time to sell all of its stock and replenish with
new
Cost-related KPIs: Admin, Inventory and Transportation
Service-related KPIs: measure overall degree of customer satisfaction

Fill rate (FR) or rate of demand satisfaction: percentage oc customer

demand that is met without delays, backorders or lost sales
Perfect order rate (POR): percentage of orders met incident-free (no
inaccuracies, damages or delays)
Order-cycle time (OCT): avg time to fulfill an order since the
moment that it is placed
Example of order-cycle time

OCT of MobilTrust is made up of two components: assembly time

and transportation time
500 observations exist for assembly, and 252 for transportation

Figure: Assembly and transportation data for MobilTrust

Example of order-cycle time

For assembly X:
Sample mean X = 9.13 days
Sample std. dev. SX = 2.3 days
For transportation Y :
Sample mean Y = 9.9 days
Sample std. dev. SY = 1.55 days
OCT = X + Y :
Sample mean X + Y = 9.13
p+ 9.9 = 19.03 days
2
Sample std. dev. SX+Y = SX + SY2 = 2.77 days
Example of order-cycle time

Figure: Plot of density functions of X, Y and X + Y

Logistics management I
Four principles of business management (POLC):
Planning
Strategic planning: Decisions made ahead of time. Long time
objectives (5-year, 10-year, 20-year planning)
Tactical planning: Detailed implementation of the long-term
strategy. Medium term objectives (quarter, semester, yearly)
Operational planning: Short-term plans to meet short-term
objectives (hourly, daily, weekly)
Five areas: order processing, procurement, warehousing, inventory,
and transportation
Organizing: How the chain of responsibilities is designed
Functional model: One department per expertise
Divisional model: Some departments may be separate for different
products (.e.g. production)
Matrix model: Teams are assembled in a per project basis. Usually
large companies with short life-cycle products
Leading: Make employees identify themselves with the goals of the
company
Logistics management II

Controlling: Monitor KPIs and take corrective actions as needed

Example: Distribution at Cardena

Distribution at Cardena
Cardena is a Romanian company producing and commercializing
perforated bricks. During the eighth week of the current year, its logistics
manager noticed numerous complaints from customers due to delivery
delays. For this reason, they decided to monitor the logistics system
starting from the first week of the year, through the use of a specific
performance measure referred to as punctuality, defined as the
percentage of weekly orders delivered on time. A sufficient level of
punctuality is set by the manager to the value of 95%.
Example: Distribution at Cardena

Figure: Punctuality the first 8 weeks at Cardena

Example: Distribution at Cardena

Distribution at Cardena
The performance measure values from the sixth week to the eighth, in
fact, confirmed the manager’s perception. Consequently, they decided to
implement a series of corrective actions to improve fleet size and vehicle
routing
Example: Distribution at Cardena

Result
The corrective measures resulted in far more orders being met at
punctuality. In particular, the goals were achieved for 6 weeks in a row
for weeks 20-25

Figure: Weekly observations measuring punctuality

Monitoring multiple performance measures

Sometimes we may identify several KPIs impacting the efficiency of

a SC
Idea: group the different KPIs in groups, normalize their values,
assign weights to each of them
Establish a threshold of desirable performance
Monitor the different KPIs simultaneously, and take corrective
actions if needed on each one of them
Monitoring multiple performance measures
Borg
Borg is a Canadian company producing wooden utensils. Following a
recent organizational restructuring, the new logistics manager is in charge
of monitoring the most critical supply chain activities every month. When
designing a control panel, it was found that the most significant problems
were connected to the large number of complaints received about errors
in dispatched orders, frequent delivery delays, an incorrect policy of
inventory management, overstaffing in the warehouse and inefficiencies in
the transportation system.

Borg: KPIs
The logistics manager therefore identified 19 performance measures,
subdivided into five families calculated with a monthly frequency: two of
them (storage and delivery) are representative of typical logistics
activities; the others (order dispatch, etc.) are defined considering the
Borg specific needs. These measures are described in the following Table,
in which the calculation method used for each of them is also indicated.
Monitoring multiple performance measures

Figure: Example: Multiple performance measures aggregated in groups

Monitoring multiple performance measures

Figure: Normalized and weighted values for each KPI

Monitoring multiple performance measures

Desirable performance
The logistics manager has determined that the minimum value to be
achieved for each performance measure should be 6, and the objective
should be 10. The control panel was constructed using a radar chart
which has a great visual impact. A control panel has been built to
visualize the performance on each category and identify the areas needing
priority corrective action (those with a value lower than 6).
Monitoring multiple performance measures

Figure: Radar chart for visualizing multiple performance measures

Data analytics

Historically, managers have taken decisions based on experience

Data flow of logistics operations can sometimes be very large

Logistics analytics
Logistics analytics is the discipline that uses data, along with information
retrieval, statistical analysis, mathematical optimization, and simulation
models, to help managers analyze and coordinate logistics systems in
order to ensure the smooth running of operations in a timely and
cost-effective manner
Data analytics

Figure: Taxonomy of analytics models

Prescriptive analytics

Prescriptive analytics
Prescriptive analytics suggests feasible actions (i.e., actions satisfying
budget, logical, temporal and technological constraints) that optimize a
given outcome z, based on available data
Example: Timor grocery wholesaler

Prescriptive analytics wirh deterministic data

Timor is a grocery wholesaler that purchases, at the beginning of every
week, a number of packs, typically between 1600 and 2000, of organically
grown salad, which is a highly perishable product, at e 1.00 each. During
the week, the company distributes the salad packs to some retailers, at
e 1.50 each, on the basis of received orders. At the end of the week, the
unsold packs are discarded since they have deteriorated. The problem
faced by Timor is to determine how many salad packs to purchase in
order to maximize its profit
Example: Timor grocery wholesaler

Prescriptive analytics wirh deterministic data

There are five alternatives shown in the Table below. The profit (in e )
for each choice is reported in the third column of the same table under
the hypothesis that next week the retailers’ orders will amount to 1780
packs

Figure: Profit (in e ) for the five alternatives faced by Timor

Example: Timor grocery wholesaler

Prescriptive analytics wirh deterministic data

The best alternative in this case is to buy 1800 packs of salad
Example: Timor grocery wholesaler
Bayes criterion
Suppose that now the demand of salads is a random variable Θ. The
purchase decision must be taken a priori, before the exact amount of
demand is revealed. If the same process is repeated every week, we have
interest in choosing the alternative that maximizes the expected profit

Discrete case, realizations θ1 , . . . , θk of the demand with

probabilities p1 , . . . , pk . For each demand realization θj and
purchase decision i, we can compute the resulting profit zij . The
expected profit associated with a purchase decision i is:
k
X
EΘ (P rof iti ) = pj zij
j=1

Continuous case, density function pΘ (θ), punctual profit of decision

i given demand realization θ given by zi (θ):
Z +∞
EΘ (P rof iti ) = pΘ (θ)zi (θ)dθ‘
−∞
Example: Timor grocery wholesaler

Bayes criterion
Let us now assume that the number of salad packs ordered by the
retailers at the beginning of the week can be modeled as a discrete
random variable θ, whose realizations are 1600, 1700, 1800, 1900 and
2000, with estimated probabilities P (θ = 1600) = 0.1, P (θ = 1700) =
0.2, P (θ = 1800) = 0.3, P (θ = 1900) = 0.2, P (θ = 2000) = 0.2
Example: Timor grocery wholesaler

Figure: Expected profits per alternative

Example: Timor grocery wholesaler

Bayes criterion
The best alternative is i = 3, or to purchase 1,800 packs of salad, for an
expected profit of e 840
Indifference zone selection

Indifference zone selection

Often the probability distributions are unknown, but an estimate can be
provided from a set of m samples. If Z denotes the sample mean and
EΘ (Z) the actual expectation, for a given confidence level 1 − α, we can
estimate
S S
P (Z − tα/2,m−1 √ ≤ EΘ (Z) ≤ Z + tα/2,m−1 √ ) = 1 − α,
m m

where tα/2,m−1 is the quantile of order 1 − α/2 of a t-student dist with

qP
(Zi −Z)2
m − 1 degrees of freedom and S = m−1
Indifference zone selection

Indifference zone selection

Suppose that we have two alternatives, A and B for a given policy. We
can compute ZA , ZB and confidence intervals for the parameters
obtained for the two alternatives. For a given indifference parameter δ, if
|ZA − ZB | > δ we can discriminate between the two and select one over
the other
Indifference zone selection

Indifference zone selection

If not, we estimate the number of additional samples that we would need
to add to alternatives A and B to bring the alternatives farther than δ.
This can be achieved using the RINOTT procedure. The RINOTT
procedure returns a real number r which we use to compute the enlarged
sample size as " 2 #
rSi
µi =
δ
We enlarge our samples for ZA , ZB and recompute the averages for the
increased samples. We then select the one with a better value of Zi