Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Introduction to the Course

4DM20 Engineering Optimization

Mauro Salazar and Dinesh Krishnamoorthy

Department of Mechanical Engineering

Eindhoven University of Technology

Academic year 2023–2024

1 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

4DM20 Team

Mauro Salazar, Dinesh Krishnamoorthy,

Responsible Lecturer Lecturer

Finn Vehlhaber, Leonardo Pedroso, Lotte Hollander, Maedeh Izadi, Janne Daams,

Responsible TA Responsible TA TA TA TA
2 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

What is Optimization?

Merriam-Webster: “An act, process, or methodology of making

something (such as a design, system, or decision) as fully perfect,
functional, or effective as possible.
specifically : the mathematical procedures (such as finding the
maximum of a function) involved in this.”

Cambridge Dictionary: “The process of making something as good

or effective as possible.”

Oxford Reference: “The process of finding the best possible solution

to a problem. In mathematics, this often consists of maximizing or
minimizing the value of a certain function, perhaps subject to given
constraints.”

3 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Example: slider-crank mechanism

5 4 3 2 1

x2 b4 P 6 8
y2
θ2 b7 2 7
3
1 b2
b1
∆γ

b5 b8 b3

b6

Find dimensions bi such that, for one full rotation of the input
crank, point P follows the predefined path as close as possible

(Paradis and Wilmert, 1983)

4 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Example: nonlinear parameter fitting

Model:
(a; t) = a1 + ta2 + e a3 t

y
Measurements:
(tj , yj ), j = 1, . . . , m

Sum of squared residuals:

Pm
t1 t2 t3 t4 t5 t SS(a) = [yj (a; tj )]2
j=1

Find parameters a such that the best model fit is obtained.

(Nocedal and Wright, 2006)

5 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Example: Pandemic Control (NLP)

Covid-19 vaccine allocations minimizing infections or fatalities
2

⇥105 Vaccine prioritization per group

⇥105 Vaccine prioritization per group
12-19 1st 12-19 2nd
12-19 1st 12-19 2nd
2.5 20-29 1st 20-29 2nd
2.5 20-29 1st 20-29 2nd
30-49 1st 30-49 2nd
30-49 1st 30-49 2nd
2.0 50-59 1st 50-59 2nd
Number of doses

2.0 50-59 1st 50-59 2nd

Number of doses
60-69 1st 60-69 2nd
60-69 1st 60-69 2nd
1.5 70-79 1st 70-79 2nd 1.5 70-79 1st 70-79 2nd
80+ 1st 80+ 2nd 80+ 1st 80+ 2nd
1.0 1.0

0.5 0.5

0.0 0.0
⇥105 Daily prevalence and fatalities ⇥105 Daily prevalence and fatalities
2.0 Not vaccinated 2.0 Not vaccinated
60 60
Prevalence (#)

Prevalence (#)
Partially vaccinated

Fatalities (#)
Partially vaccinated

Fatalities (#)
1.5 Fully vaccinated 1.5 Fully vaccinated
40 40
1.0 1.0
20 0.5 20
0.5

0.0 0 0.0 0

Disease progress Disease progress

1.0 1.0

0.8
0.8
Share of population
Share of population

0.6
0.6

0.4
0.4

0.2
0.2

0.0
2021 02 2021 03 2021 04 2021 05 2021 06 2021 07 2021 08 2021 09
0.0
2021 02 2021 03 2021 04 2021 05 2021 06 2021 07 2021 08 2021 09 Date
Date E R v2 R R S v1 R v2 I v1 R v2 I v2
E R v2 R R S v1 R v2 I v1 R v2 I v2 I R v1 R S R v1 I v1 S v2
I R v1 R S R v1 I v1 S v2

S. Tonkens, P. de Klaver and M. Salazar, “Optimizing Vaccine Allocation Strategies in Pandemic

Outbreaks: An Optimal Control Approach”, ECC, 2022
6 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Example: Electric Racing (SOCP)

Minimum-lap-time design and operation for e-racing

O.J.T. Borsboom, C.A. Fahdzyana, T. Hofman and M. Salazar, “A Convex Optimization Framework for
Minimum Lap Time Design and Control of Electric Race Cars”, IEEE TVT, 2021
7 / 17
 the first study that Definition
A. Recent Research analyzes the benefitand Examples of such an Objectives
unlikelihood of Prerequisites
theinter- an intermodal system Topics being Materials
globally Planning Exam
here comesmodal the transportation
literature, bla bla system
bla -from a mesoscopic
somebody has to point controlled,of we derive a (Pigovian) pricing scheme that would
do it view. We develop an optimization approach that influence
finds the selfish actors to behave according to the social

Example: Intermodal Autonomous Mobility-on-Demand (LP)

B. Aims and
optimal control policy for this system under steady
Scope
conditions. Herein, we incorporate different objectives that
optimum
state [Link] Multi III-B.
Commodity Flow Based Optimization Approach
As can beconsider
seen, noeither studythe ontotal transportation
centrally time, or the generated
operated intermodal
emissions, orexists
passenger transportation both so byfar,
incorporating
especially with a convexly
respect combined

Find the users’ intermodal routes that minimize societal cost

to AMoD objective
systems. as
a case
the first study thatstudy
well asthis
Against
based on
analyzes
a generalized
background,
thereal-world
benefit ofdata
costwefunction.
such from
provideWe provide
an Manhattan.
inter-
To represent the transportation system and its different
Based transportation modes, we use the (in)complete layered graph
on the results
modal transportation for this
system fromstudy, we derive point
a mesoscopic managerial
of insights G = (V , A ) shown in Fig. 1 with a set of vertices V [FR]:?
A. Multi Commodity Flow Based Optimization Approach fm (i, j) and rebalancing flows f0 (i, j)
for bothan
view. We develop fleet operators and
optimization municipalities.
approach that finds the andthe a set of arcs Asystem V ,itscomprising
optimal control Thepolicycontribution
for thisofsystemour study underis fourfold:
steady state First, we Topro- represent transportation ✓ V ⇥and different aconstraint road network (1b) guarantees flow conservat
A. Multi Commodity
transportation layer
modes, GRwe =Flow
(VRthe
use ,Based R ), aOptimization
Adigraph subway
G = (V , layer Approach
A ) shown GS = (V S , AS ), 1and
whereby j=x is a a boolean indicator fu
conditions. vide
Herein, the we first optimization
incorporate framework
different objectives for that
an intermodal pedestrian layer G = (V , A ). The road layer
in Fig. 1, which has a set of PverticesP V Pand a set of furtherrepresents flow conservation for vehicles in E
autonomous
consider either mobility-on-demand
the total transportation time, or(I-AMoD)
the generated system,arcs which A ✓ Vintersections
⇥ V . The graph i 2 VRcontains and road (i, j) 2 ARcapacity
links network
a road . The subway limits for roads in Eq. (1d) and pu
emissions, handles
or both real-world
by incorporating data sets in short combined
a convexly computationallayer times GR = (V layer R ), a subway
R , Acomprises subwaylayer stops GP = (V i 2P ,VASP ),and links in [Link]
andthe respective (1e).
[MaS]:public
objective asand welldelivers global optimality.
as a generalized Second,
cost function. Wewe provide
provide a sound
trans- a pedestrian
To represent layer =, (V W , AWthe
2GWAtransportation
j) the ). Herein, theand road layer
(i, S while pedestrian
system layer its represents
differentB. I-AMoD walkable Objective
a case studycasebasedthat is based on
on real-world datareal-world data for
from Manhattan. Manhattan,
Basedporta- represents an
transportation intersections i 2 VR and road links (i, j) 2 AR .
streets
modes, (i, j)we2use AP the in (in)complete
between intersections layered graph i 2The VP .generalized
Finally, cost function (1a) can
tion? The subway layer comprises subway stops i 2 VP connected
urban
on the results forareathis in which
study, wethederive
need managerial
for a sustainable insightstransportation
Gby =arcs (V (i,, Aj)arcs
)2shown out of in set Fig.
A ✓1VRwith ⇥VPalayer[V setS ⇥V connectdifferent
of Pvertices the
[FR]:?
V pedestrian
concept
for both fleet operators is more than urgent. Third, we present results
and municipalities. that AP , while theC pedestrian represents objectives. In our studies, we o
and
walkable oflayer
a setstreets arcs(i,toA2the
j) ✓AV road
⇥ V ,and
between to the subway
comprising
intersections a road
i 2 V layer,
network
. respectively,
welfare by minimizing overall costs. Spe
are not limited
The contribution of our study to a single objective
is fourfold: First,but weinclude
pro- different
layer = such that
W
a C=subway
W
Finally,GRarcs (V
outR ,ofA ), VA
Rset ✓VVPR[⇥Vlayer VRW[[VVSGP,S⇥ A=VW = connect
(V SA,A P[ ARand
S ), AaS [ AC costs
[commuting and that depend on the cu
perspectives:
vide the first optimization i) theframework
social welfare for inanmonetary
intermodal terms of value
pedestrian layer 0/ (Vholds. ). The road layerlayer, represents time VT and on operational costs for the A
the pedestrian \ VGSPto
VRlayer ==the road
P , APand to the subway
autonomousofmobility-on-demand
time and operational(I-AMoD) costs, andsystem,
ii) the which
social welfare in
intersections i 2 VRarcs andmodelroad links (i, j) 2 Aability subway. Herein, costs for the AMoD flee
respectively. These the customer’s R . The tosubway
both monetary
handles real-world data sets andinenvironmental terms. Fourth,
short computational times welayer derive
switch transportation
comprises Wesubway
use thestops
modes, following
suchi 2 that VSVnotation
and= Vthe to
VRdescribe
W [respective[ VP , lines characteristics
dependent ownership costs VD,R to accou
[MaS]:Update
and deliversmanagerial
global optimality. insights Second,
that, besides providing
we provide dedicated
a sound A j)
(i, intu-
= 2WA[SA
A ,ofRwhile
[GAandP[ Adefine
the and Vour
Cpedestrian = 0/ holds.
R \ Voptimization
Player represents problem: walkable and depreciation
Each arc has a as well as energy costs V
case that isitionsbasedforonsingle stakeholders,
real-world data foranalyse
Manhattan, the social
an optimum To consider congestion we use a simplified threshold system, V comprises all operational c
streets (i, j)capacity 2 AP incibetween j which denotes intersections eitheri 2 theVPcapacity
. Finally, of a certain D,P
kilometer. This way, we define the social
urban area inthat can be
which thereached.
need for a sustainable transportation arcs model: Eachtransportation
arc (i, j) has mean a capacity (AR , PA ci j )which denotes as ci j = •, 8(i, j) 2
out of set flow S or remains
connect
concept is more Thethan remainder of thiswepaper
urgent. Third, is structured
present the maximum
results thatas follows:
AC ✓ofVpassengers
R ⇥VP [VSor⇥V vehicles that the the pedestrian
arc
layer to theACroad , APwithoutfor transportation
and toencountering
the subway means
layer, without
respectively, capacity (i, j) , f0 (i, j)) = VT · Â ti j · fm (i
CM ( fm limits,
can accommodate traffic congestion
Section
are not limited to aII single
presents the methodological
objective but include background
different such for our i.e., walking.
VP [ VR [ The travel =AAPtime AtRicapacity
j [denotes ACtheand average time m2M ,(i, j)2A
((i, j) that
2 A V) or=overcrowding , A
VS((i, j) 2 P[
). The AS [of
perspectives:studies. Section
i) the social III derives
welfare a pricing
in monetary terms scheme
of value to steer self- R needed
Vwalking
R \ VS arcs holds. to
= 0/ remains as traverse
ci j = •, 8(i, an j)arc 2A (i, j). Times on arcs (i, j) 2 AC
C , AW . Travers- + Â(VD,R · di j +VE · eR,i j ) · f0 (i, j) +
of time andinterested
operational agents
costs,to andthe ii)
social optimum.
the social Then,in Section
welfare IV represent switching times
ing an arc (i, j) takes on average ti j time units. Note herein, between or to reach a certain mean
(i, j)2AR
both monetarydetails andour case study terms.
environmental and discusses
Fourth, we our derive
experiments We
that and
ti j use
8(i, j) of2 transportation.
the following
AC denotesnotation the Let timeR be
tonecessarythe set
describe of all travel +V
tocharacteristics
switch requests. A
[MaS]:Update D,P · Â di j · Â f m (i, j) .
managerial results.
insightsFinally, Sectionproviding
that, besides V concludes the paper
dedicated intu-withof a short
between
G andtwo meansour
request
define of transportation.
= (om , dm , amproblem:
rmoptimization Given
) 2 RtheisEach threshold
a triple arc composed
has a (i,by an
j)2A P m2M
summary
itions for single and an outlook
stakeholders, analyse on the
future research.
social optimum capacity modeling capproach,
i jorigin
whichnode we assume
denotes tidestination
om , aeither j to be the constantnodeif danmofarc’s
capacity and a request rate am
a certain
that can be reached. capacity constraint holds. Given the mesoscopic nature of our stud
II. M ETHODOLOGY transportation thatmean denotes (AR ,the AS )amountor remains as ci j = •,per
of customers 8(i,unit
j) 2time.
energy
Since of a single vehicle e
consumption
The remainder of this paper is structured as follows: AC , AP forwe Let R be the set of all travel requests. A request rm = R
identify different
transportation meanstransportationwithout capacity modeslimits, by different
assuming thatarc road arcs are traversed at
This section presents the methodological m , am ) 2 R is a triple of an origin node om 2 VW ,
(om , dfor
Section II presents the methodological background for background our i.e., walking. sets, The wedtravel
use only
time ati ajsingle
denotes type theofaaverage
flow variables
time di j f (i, j)
our studies. We aaim at analyzing a destination node 2 V , and request rate that v = . Considering
m electric vehicles wi
scheme tothe benefit
self- of needed
AMoD m W m i j t
studies. Section III derives pricing steer denotes to that
the traverse
amount denotes
of arcthe(i,flow
an customers j).per onunit
Times an on arcarcs
time (i, j)
for (i, for
eachj) 2acapabilities
Acertain
C
ij
travel
and an overall tank-to-wheel
systemsto intheansocial
interested agents intermodal
optimum. setting.
Then,Herein,
Section weIV use a fluidic
represent
request. Note request
that om m
switching times
and
2M dmbetween
lie
= [1,on M] theor✓ [Link]
pedestrianFurthermore,
adigraph.
certain mean fenergy
0 (i, j) consumption
denotes for a road arc is
details our optimization
case study and approach to determine
discusses the optimal
our experiments andequilibrium
Accounting
of transportation. for different
the Let Rtransportation
rebalancing beflowthe set of emptymodes
of by separate
AMoD
all travel vehiclesAon the⇣ road
requests.
for such a system. Within this approach, we consider r ⌘ d
results. Finally, Section V concludes the paper with a short request arc sets, rfmm = (i,
arcs j) m
(o denotes
, dj)
(i, m ,2am Athe
)R2. flow
R is on aarc (i, j)composed
triple 2 A for a by an eR,i j =
a
· Af · cd · v2i j + cr · mv · g ·
i

summary and •anthe assignment

onM. of transportation
Salazar,
research. N. Lanzetti, requests to F. transport
Rossi,
certain travel M. Schiffer
m 2 M =and [1, M]M. ✓ [Link], 2 hE
outlook future origin node orequest
m , a destination node dm and a request rate am
To account for
flows, “Intermodal Autonomous Mobility-on-Demand”, rebalancing flowsWith between a customer’s
this notation, destination and the
II. MmodesETHODOLOGY that denotes the amount of IEEEcustomers
next customer’s origin, f0 (i, j) denotes the flow of empty
the I-AMoD
T-ITS,
per unit2019 optimization
time. Since problem
different of transportation, holds as follows: For a given setby of different
transportation demands
8 / 17 This section• presents the methodological background for
we identify
vehicles on road different
arcs (i,transportation
j) 2 AR . modes arc first
The term in (3) represents the aerod
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Example: Electric Aircraft Routing (MILP)

Assign e-aircraft to flights and charging schedules such that the
curtailment of renewable energy is reduced
(a, t + 3) (b, t + 3)
C(i,j )
(a, t + 2) (b, t + 2)
Pgr
(a, t + 1) (b, t + 1)
E b Pb Pa (i, j )
BESS (a, t) (b, t)
Prnw apron t

F. Vehlhaber and M. Salazar, “Electric Aircraft Assignment, Routing, and Charge Scheduling
Considering the Availability of Renewable Energy”, IEEE L-CSS 7, 2023

9 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Course objectives

After successful completion of the course, the participant

• knows the basic concepts and terminology of engineering
optimization;
• is able to formulate an optimization problem;
• has background and working knowledge of algorithms for
numerical optimization, gradient-based methods in particular;
• is able to solve optimization problems using the Matlab
optimization toolbox and YALMIP, and to interpret the results;
• is able to formulate and solve data fitting problems using the
Matlab optimization toolbox;
• understands the principles of parallel computing methods and
tools for large-scale data fitting problems;
• is able to read text books on engineering optimization.

10 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Prerequisite knowledge

• Calculus and vector-calculus

• Linear algebra
• Some Matlab experience
• Elementary knowledge of numerical methods
• (Elementary knowledge of mechanics, kinematics, dynamics,...)

11 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Topics treated in the course

• Optimization problem formulation

• Convex sets, functions and optimization
• Conditions for optimality
• Gradient-based algorithms for interior optima and data fitting
• Gradient-based algorithms for boundary optima
• Use of algorithms available from the Matlab Optimization
toolbox and YALMIP to solve engineering optimization
problems

12 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Course materials

• Book: Convex Optimization by Boyd and Vandenberghe,

Cambridge University Press (2004), PDF available online;
• Book: Numerical Optimization by Nocedal and Wright,
Springer, second edition (2006);
• Book: Principles of Optimal Design by Papalambros and
Wilde, third edition (2017);
• Matlab optimization toolbox
• YALMIP and some numerical solvers
• Slides and supplementary material (see Canvas)
• Exercises and study questions (see Canvas)
• Computer and self-study assignments (see Canvas)

13 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Course planning

Set-up:
• Mondays, hours 5 – 8: Lectures with exercises
• Wednesdays, hours 1 – 2: Guided self-study in presence

Format:
• The lectures with exercises are given in presence, in Gemini
Zuid College Zaal.
• The guided self-study sessions are organized in Gemini Zuid
3A08. You can work with others on the computer assignments,
and ask questions to the TAs.

On videocollege you will find video recordings from the 2022 edition
and from 2023 D. Krishnamoorthy’s new lectures. Yet we strongly
encourage you to join the lecture in presence.

14 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Course planning
Week Date Subject
6 Feb 5, 7 L: Introduction & Problem formulation
A: Optimization with Matlab
7 Carnival break
8 Feb 19, 21 L: Convex optimization & Conditions for optimality
A: Optimization with YALMIP
9 Feb 26, Feb 28 L: Conditions for optimality for constrained problems
A: Optimization with Matlab
10 Mar 4, 6 L: Unconstrained optimization
A: Optimization with Matlab
11 Mar 11, 13 L: Unconstrained optimization & Data fitting
A: Data fitting
12 Mar 18, 20 L: Constrained optimization
A: Data fitting & Distributed data fitting
13 Mar 25, 27 L: Constrained & Distributed optimization
A: Distributed data fitting
14 Apr 3 Tutorials and Back-up
back-up
15 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Written Exam

• The exam will be written and based on this year’s material

• Further details will be announced in due time

16 / 17
 Definition and Examples Objectives Prerequisites Topics Materials Planning Exam

Best Teacher Award

Vote for your

favorite teacher
and win a €20,-
[Link] voucher!
17 / 17