Available

Available online at www.sciencedirect.com

Available online
online at
at www.sciencedirect.com
www.sciencedirect.com
Available online online
Available at www.sciencedirect.com
at www.sciencedirect.com

ScienceDirect
Procedia Manufacturing 00 (2019) 000–000
Procedia
Procedia Manufacturing
Manufacturing 00
00 (2019)
(2019) 000–000
000–000 www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
Procedia Manufacturing
Procedia 42 (2020)
Manufacturing 475–482
00 (2019) 000–000 www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
International
International Conference on Industry 4.0 and Smart Manufacturing (ISM 2019)
International Conference
Conference on
on Industry
Industry 4.0
4.0 and
and Smart
Smart Manufacturing
Manufacturing (ISM
(ISM 2019)
2019)
International Conference on Industry 4.0 and Smart Manufacturing (ISM 2019)
Managing
Managing loading
loading and
and discharging
discharging operations
operations atat cross-docking
cross-docking terminals
terminals
Managing loading and discharging
M. Flavia Monacoa,∗
operations at cross-docking
a,∗, Marcello Sammarrabb
terminals
M. Flavia
M. Flavia Monaco
Monacoa,∗,, Marcello
Marcello Sammarra
Sammarrab
a,∗ bVia P. Bucci 44E, 87036 Rende (CS), Italy
M. Flavia Monaco , Marcello Sammarra
aaDipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica,
a Dipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica, Università
a Dipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica, Università
bbIstituto di Calcolo e Reti ad Alte Prestazioni, Consiglio Nazionale delle
della Calabria,
Università della
della Calabria,
Calabria, Via
Via P. Bucci
P.87036 44E,
44E, 87036
BucciRende 87036 Rende
Rende (CS),
(CS), Italy
Italy
b Istituto di Calcolo e Reti ad Alte Prestazioni, Consiglio Nazionale delle Ricerche ,, Via P. Bucci 7-8C, (CS), Italy
a b Istituto di Calcolo e Reti ad Alte Prestazioni, Consiglio Nazionale delle Ricerche , Via P. Bucci 7-8C,
Ricerche Via P. Bucci 7-8C, 87036
87036 Rende
Rende (CS),
(CS), Italy
Italy
Dipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica, Università della Calabria, Via P. Bucci 44E, 87036 Rende (CS), Italy
b Istituto di Calcolo e Reti ad Alte Prestazioni, Consiglio Nazionale delle Ricerche , Via P. Bucci 7-8C, 87036 Rende (CS), Italy

Abstract
Abstract
Abstract
A cross-docking terminal
Abstract
A isis aa relevant node within a distribution chain. Actually, at this intermediate logistic platform between suppliers and
A cross-docking
cross-docking
retailers, incoming
terminal
terminal
flows of a relevant
ispossible
relevant node
different
within
within aa distribution
nodecommodities distribution
are
chain.
chain. Actually,
consolidated Actually,
into
at this
this intermediate
at shipments,
single intermediate
with
logistic
logistic
respect to
platform
platform
the retailers’
between
between suppliers
suppliers
orders, and
and
and
directly
retailers,
retailers,
A incomingterminal
incoming
cross-docking flows of
flows ofispossible
possible
a relevantdifferent
different
node commodities
commodities
within a are consolidated
are consolidated
distribution chain. into single
into single
Actually, at shipments,
shipments,
this with respect
with
intermediate respect to the
to
logistic the retailers’
retailers’
platform orders,suppliers
orders,
between and directly
and directly
and
delivered,
delivered, skipping
skipping thus
thus the
the storage
storage phase.
phase. In
In such
such a context
aa context the
the synchronization
synchronization of
of the
the inbound
inbound and
and outbound
outbound trucks
trucks is
is a necessary
aa necessary condition
condition to
to
delivered,
retailers,
guarantee skipping
incoming
fast and thus
flows the
of
congestion-freestorage
possible phase.
different
transshipmentIn such context
commodities
operations. are the
In synchronization
consolidated
this paper weinto of
single
propose the inbound
shipments,
a Mixed and
with
Integeroutbound
respect
Linear totrucks
the
Program is
retailers’
and necessary
a orders, condition
heuristicand to
directly
algorithm
guarantee fast
guarantee
delivered, fast and congestion-free
and
skipping congestion-free
thus the storage transshipment
transshipment
phase. In suchoperations.
operations.
a context In this
In
the this paper we
paper we propose
synchronization propose
of thea inbound
a Mixed Integer
Mixed Integer
and Linear Program
Linear
outbound Program
trucks isand
and
a a heuristic
a heuristic
necessary algorithm
algorithm
condition to
for
for managing the loading and discharging operations, with the aim of minimizing the completion time of the whole transshipment process.
for managing
managing
guarantee fastthe
andloading
the loading and
and discharging
congestion-free discharging operations,
operations,
transshipment with
with the
operations. theInaim
aim thisof minimizing
ofpaper
minimizing
we proposethe
the completion
completion time
time of
a Mixed Integer the
the whole
ofLinearwhole
Program transshipment
transshipment process.
process.
and a heuristic algorithm
for managing the loading and discharging operations, with the aim of minimizing the completion time of the whole transshipment process.
©
 2020
cc 2020
2020 The TheAuthors.
The Authors.
Authors. Published
Published
Published by by Elsevier
by Elsevier
Elsevier B.V.B.V.
B.V.
c 2020 The Authors. Published by Elsevier B.V.
This
This
This isan
isis an
an open
open
open access
access
access article
article
article under
underunder
the
the theBY-NC-ND
CC
CC CC BY-NC-ND
BY-NC-ND licenselicense
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
c 2020
This is anThe open access
Authors. article under
Published thescientific
by the CC BY-NC-ND
Elsevier B.V.committee license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review
Peer-review
Peer-review under
under
under responsibility
responsibility
responsibility of
of of
thethe scientific
scientific committee
committee of
of theof
the the International
International
International Conference
Conference
Conference on on Industry
on Industry
Industry 4.0 and4.0
4.0 and and Manufacturing.
Smart
Smart Smart Manufacturing.
Manufacturing.
Peer-review
This is an open under responsibility
access article under of the scientific
CC BY-NC-NDcommittee of the
license International Conference on Industry
(http://creativecommons.org/licenses/by-nc-nd/4.0/)4.0 and Smart Manufacturing.
Keywords:
Peer-review under responsibility
Keywords: scheduling;
scheduling; of the scientific committee of the International Conference on Industry 4.0 and Smart Manufacturing.
heuristic; logistics
Keywords: scheduling; heuristic;
heuristic; logistics
logistics
Keywords: scheduling; heuristic; logistics
planning,
planning, where where the the objective
objective of of reducing
reducing as as much
much as as possible
possible
1. Introduction and related works planning,
the inventory where levels the isobjective
pursued. of reducing as much as possible
1.
1. Introduction
Introduction and and related
related works works the inventory
planning,
theTheinventory where levels
levels is pursued.
the isobjective
pursued. of reducing as much as possible
scientific literature on cross-docking problems is very
1. Introduction and related works theThe scientific
inventory
The scientific levels literature
is pursued.
literature on
on cross-docking
cross-docking problems
problems is
is very
very
In
In the
the last
last twotwo decades,
decades, cross-docking
cross-docking strategies strategies have wide,
have gained
wide,
gained and
and also a summary
also aa literature
summaryonanalysis analysis
analysis goes behind
goes behind
behind the
the isaim
aimvery of
of
In the last two decades, cross-docking strategies have wide,
gainedTheand scientific
also summary cross-docking goes problems the aim of
aa key
key role
role in
in logistics
logistics industries.
industries. As
As underlined
underlined in
in [9],
[9], this
nowadays
this
nowadays paper.
paper. We
We refer
refer the
the interested
interested reader
reader to
to the
the review
review papers
papers
a keyIn the
role last
in two decades,
logistics cross-docking
industries. As underlined strategies
in have
[9], wide,
gained
this
nowadays and We
paper. alsorefer a summary the analysis
interested readergoestobehindthe review the aim papers of
customers, in aa broad sense, have become even more [2, 11,
volatile
[2, 11, 3,
3, 8].
8]. We
We just
just want
want to
to underline
underline that,
that, among all the prob-
customers,
key role inin
acustomers, in a broad
logistics
broad sense,
industries.
sense, have
have become
As underlined
become even more
in [9],
even more volatile
this
nowadays
[2,
volatile paper.
11, 3, 8].We We refer
just the interested
want to underline reader that, toamong
the review
among all
all the
the prob-
papers
prob-
and
and impatient than before. This high pressure lems
by customers,lems arising
arising in
in the
the management
management of
of aaa cross-docking
cross-docking terminal,
terminal, in
in
and impatient
customers,
calling impatient
for
in athan
fast
broad
than
deliveries
before.
sense,This
before.of This
their
havehigh become
high
orders,
pressure
pressure
is
evenbyby
completely
morecustomers,
[2,
volatile
lems
customers,
this
dumped
11,arising
3, 8]. We
paper we inare
justmanagement
the want to underline
concerned with ofone of
that, among all
cross-docking
the most relevant
the prob-
terminal, ones in
calling
and
calling for fast
impatient
for fast than
deliveries
before.
deliveries of their
of their orders,
This orders,
high pressureis completely
is completely
by customers,this
dumped
lems
this
dumped paper
arising
paper we
we are
inare concerned
theconcerned
management with
with one
ofone of the
a cross-docking
of the most
most relevant
terminal,
relevant ones
ones in
on suppliers and manufacturers, in general. However, at the
distribu-
at the operational
operational decision
decision level,
level, that is the synchronization of
on
on suppliers
calling for fast
suppliers and
and manufacturers,
deliveries of their in
manufacturers, general.
orders,
in general. However,
is completely
However, distribu-
this
dumped
at the
distribu- paper we are concerned
operational decision with that
level, one of
that is
is the
the synchronization
most relevant ones
synchronization of
of
tion processes based on cross-docking strategies present inbound
bene-
inbound and
and outbound trucks, also known as truck scheduling
tion
on
tion processes
suppliers
processes based
and on cross-docking
manufacturers,
based on cross-docking in strategies
general. However,
strategies present
present bene-
at
distribu-
inbound
bene-
problem. and outbound
the operational outbound decision trucks,
trucks, level,also
also that known
is theas
known as truck
truck scheduling
synchronizationscheduling of
fits both
fits both
both for customers
for customers
customers and
and suppliers.
suppliers. strategies present bene- problem.
tion
fits processes
for based onand cross-docking
suppliers. inbound
problem. and outbound trucks, also known as truck scheduling
A cross-docking terminal is a logistic platform, where Generally,
prod-Generally, truck
truck scheduling
scheduling problems
problems can
can be divided into
fitsA cross-docking
Aboth for customers
cross-docking terminal
terminal is
is aa logistic
and suppliers. logistic platform,
platform, where
where prod-
problem.
Generally,
prod-
two main classes: truck scheduling
scheduling of problems
inbound can be
trucks,
divided
bewhile
divided it is
into
into
as-
ucts
ucts arriving
arriving by
by inbound
inbound trucks,
trucks, are
are arranged
arranged with
with respect
two
respect to
main
to classes: scheduling of inbound trucks, while it is as-
ucts A cross-docking
arriving by terminal
inbound is
trucks, a logistic
are platform,
arranged with where
respect Generally,
prod-
two main
to classes:truck scheduling
scheduling of problems
inbound can
trucks, be divided
while it isinto
as-
customers’
customers’ requirements
requirements and
and loaded
loaded into
into the outbound
the outbound
outbound sumed
trucks,
sumed
trucks, that the
thatclasses: outbound
the outbound
outbound trucks
trucks are already
are already
already scheduled
scheduled or
or as-
as-
ucts arrivingrequirements
customers’ by inbound and trucks, loaded are intoarranged
the with respect two
sumed
trucks, main
to that the scheduling trucksof inbound
are trucks, while itor
scheduled is as-
possibly skipping the storage phase or at most being signed
stored for
signed on
on a midterm horizon; scheduling of inbound and out-
possibly
customers’
possibly skipping the
requirements
skipping the storage
and phase
storage loadedor
phase or at
into
at most being
the outbound
most being stored for
sumed
trucks, thataa the
signedtrucks.
stored for on midterm
midtermoutbound horizon;
horizon; trucks scheduling
are already
scheduling of
of inbound
scheduled
inbound and
and orout-
as-
out-
aa short amount of time (typically less than 24 hours). bound
By re-
bound trucks. Our
Our problem
problem belongs
belongs to
to the
the second
second class.
class. There-
There-
a short
possibly
shortthe amount
skipping
amount of
ofthe time
time (typically
storage
(typicallyphaseless or atthan
less most
than 24 hours).
24abeing
hours). By
stored
By
fore,
re-
signed
for
boundre- on
trucks.a midterm
Our horizon;
problem scheduling
belongs to the of inbound
second class. and out-
There-
ducing
ducing
aducing
short thethe inventory
amountinventory
of time
costs,
costs, cross-docking
cross-docking
(typically less than
have
have positive
24aahours).
positiveBy fore, our
effect
effect
bound
fore,re-our aim
aim is
trucks.
our aim is
Our
is
the
the
coordinate
theproblem
coordinate scheduling
belongsscheduling
to the second
of
of thethe inbound
inbound
class.
and
There-and
forfor the
the
inventory
suppliers,
suppliers, given
given
costs,
that
that
cross-docking
itit allows
allows to
to
have
reduce
reduce
positive
the distributioneffect
outbound
outbound trucks
trucks at
at aa coordinate
cross-docking
cross-docking
scheduling
terminal,
terminal,
of the
with
with
inbound
the
the objective
objective
and
ducing
for the the inventory
suppliers, givencosts,
that cross-docking
it allows to have athe
reduce the distribution
positive fore,
effect
outbound
distribution our aim trucks is the
at a coordinate
cross-docking scheduling
terminal, ofwiththe the
inboundobjective and
costs. On the other hand, the absence of the inventory of minimizing
phase
of minimizing the
the time needed to complete all the required op-
costs.
for
costs.theOn the
the other
suppliers,
On othergivenhand,
hand,thatthe
the absence
it allows
absence to of the
reduce
of the inventory phase
outbound
the distribution
inventory of minimizing
phase trucks the attime
time needed
a cross-docking
needed to
to complete
terminal,all
complete all the
with
the required
the objective
required op-
op-
in in the
the distribution
distribution process
process allows
allows to
to reduce
reduce the
the customers’
customers’ erations.
lead
erations.
lead As
As in inthe[4,
[4,time7],
7], we we consider
consider aa basic
basicall layout
layout constituted
constituted
costs.
in the On the
distribution other hand,
process the
allows absence
to reduce of the
the inventory
customers’ of minimizing
phase
erations.
lead As in [4, 7], needed
we considerto completea basic the required
layout constituted op-
times, so having positive effects also on them. In some by
sense,
by two
two doors
doors (or gates), one for the discharging and one for
times,
in
times, so
so having
the distribution
having positive effects
process effects
positive allows also also
to reduceon
on them. In some
the customers’
them. In some sense,
erations.
by lead
sense,two As in(or
doors [4,gates),
(or 7], weone
gates), one for
consider
for the
the a discharging
basic layoutand
discharging and one
constituted
one for
for
cross-docking
cross-docking can
can be
be viewed
viewed as
as the
the counterpart,
counterpart, on
on the
the the
logistic
the loading
logisticloading operations.
operations. However,
However, differently
differently from
from the
the previous
previous
times, so having
cross-docking can positive
be viewedeffects as also
the on them. In
counterpart, on some
the by
sense,
the
logistictwo
loadingdoors (or gates),
operations. one for differently
However, the discharging from andprevious
the one for
side, of the well established Just in time strategy in cited
production
cited papers,
papers, the
the transshipment
transshipment flows
flows between
between inbound
inbound and
and out-
side,
side, of of the
cross-dockingthe well
well canestablished
be viewedJust
established as the
Just in
in time
time strategy
counterpart,
strategyon in production
in the the
citedloading
logistic
production
bound papers,
trucks
operations.
the
are transshipment
explicitly
However, flows
considered
differently
between
as part
from
inbound
of the and out-
the previous
decision out-
side, of the well established Just in time strategy in production boundpapers,
cited
bound trucksthe
trucks aretransshipment
are explicitly considered
explicitly considered
flows between as part
as partinbound
of the
of the decision
decision
and out-
problem,
problem, while
while in
in [4,
[4, 7] they are assumed to be known.
∗∗ Corresponding author. Tel.: +39-0984-494709 ; fax: +39-0984-494781
∗ Corresponding author. Tel.: +39-0984-494709 ; fax: +39-0984-494781
∗ Corresponding author. Tel.: +39-0984-494709 ; fax: +39-0984-494781
bound
problem,
The
trucks
whileis
paper
are [4, 7]
inorganized 7] they
explicitly they as
are
are assumed
considered
assumed
follows.
asto
In to be
be known.
part
Section
of the decision
known.
22 we detail
E-mail address: monaco@dimes.unical.it (M. Flavia Monaco). The
problem,
The paper
while
paper is
is inorganized
[4,
organized 7] they as
as follows.
are assumed
follows. In
In Section
to be known.
Section 2 wewe detail
detail
∗ E-mail
E-mail address: monaco@dimes.unical.it
address:author.
Corresponding monaco@dimes.unical.it
Tel.: +39-0984-494709
(M. Flavia
(M. Flavia Monaco).
Monaco).
; fax: +39-0984-494781 the problem and provide a mixed integer linear program. The
theThe
the problem
paperand
problem and provide aa as
is organized
provide mixed
follows.
mixed integer
integer linear
In linear program.
Sectionprogram.2 we detail The
The
E-mail address: monaco@dimes.unical.it (M. Flavia Monaco). Lagrangian
Lagrangian decomposition
decomposition technique
technique is
is discussed
discussed in
in Section
Section 3,
3,
2351-9789
2351-9789
2351-9789
cc 2020 The Authors. Published by Elsevier B.V.
 ©2020
2020The Authors.
The Published
Authors. by
PublishedElsevier
by B.V.
Elsevier B.V.
the problem
Lagrangian and provide
decomposition a mixed
technique integer
is linear
discussed program.
in Section The 3,
2351-9789
This is an c
open 2020 The
access Authors.
article Published
under the CC by Elsevier
BY-NC-ND B.V.
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
This is
This
This isisan
anan
2351-9789 
open
open
c access
open article
access
access under the
article
article under
2020responsibility
the CC BY-NC-ND
under
The Authors.ofPublished CC BY-NC-ND
the CC
by Elsevier BY-NC-ND
license license Lagrangian decomposition
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
B.V.of the International Conference on Industry 4.0 and Smart Manufacturing.
technique is discussed in Section 3,
Peer-review
Peer-review under
under responsibility of the
the scientific
scientific committee
committee of the International Conference on Industry 4.0 and Smart Manufacturing.
Peer-review
Peer-review
This is an open under
under responsibility
responsibility
access of the of
article under CCthe scientific
scientific committee
BY-NC-ND committee
of the
license of the International
International Conference on Conference
Industry 4.0 and
(http://creativecommons.org/licenses/by-nc-nd/4.0/) onSmart
Industry 4.0 and Smart Manufacturing.
Manufacturing.
10.1016/j.promfg.2020.02.045
Peer-review under responsibility of the scientific committee of the International Conference on Industry 4.0 and Smart Manufacturing.
476 First Author et al. / Procedia Manufacturing 42 (2020) 475–482
Monaco and Sammarra / Procedia Manufacturing 00 (2019) 000–000 2

that is followed by the description of the Lagrangian heuristics

in Section 4. Numerical results are discussed in Section 5. Fi- Nomenclature
nally, conclusions are drawn in Section 6.
Decision Variables
xik = 1 if the truck i ∈ I is the k−th one in the inbound
sequence, 0 otherwise
2. Mathematical Model y jh = 1 if the truck j ∈ J is the h−th one in the outbound
sequence, 0 otherwise
We consider a cross-docking terminal with one door at both zipj ≥ 0 quantity of product p ∈ P shipped from truck i ∈ I
inbound and outbound sides. A truck at a time can be processed to truck j ∈ J
at the docking doors. The products shipped from a given set of vipj = 1 if a strictly positive amount of product p ∈ P is
inbound trucks have to be unloaded at the first door, sorted on shipped from i ∈ I to j ∈ J, 0 otherwise
the basis of the customer demands, moved across the dock, and S iI ≥ 0 starting processing time of the truck i ∈ I
then loaded onto a set of outboubd trucks, at the second door. CiI ≥ 0 completion time of the truck i ∈ I
We assume, as it is usual in this context, that all trucks are ready S Oj ≥ 0 starting processing time of the truck j ∈ J
at the beginning of the planning horizon, their processing times C Oj ≥ 0 completion time of the truck j ∈ J
are known, while the transshipment time inside the terminal is
Cmax the makespan, Cmax = max j∈J {C Oj }
negligible. Moreover, preemption in the unloading/loading of
each truck is not allowed. Therefore, the processing of an out-
bound truck cannot start before the unloading of all inbound Note that the variables S iI and S Oj have been introduced for
trucks carrying loads for it has been completed. The problem is sake of readability, even though they are redundant (see con-
to decide the unloading and loading sequences of the trucks so straints (5) and (9) in the model). With the above notation, the
as to minimize the completion time of the whole process, that Mixed Integer Linear Model for the truck-scheduling problem
coincides with the time at which the last outbound truck has under investigation is the following:
been completely loaded and can leave the terminal (makespan).
Here is the list of the notation we adopt to formulate the
truck-scheduling problem.
Z = min Cmax (1)

xik = 1 i∈I (2)
Nomenclature k∈K

xik = 1 k∈K (3)
Main Notation
i∈I
I Set of inbound trucks, |I| = n S iI ≥ ClI − M I (2 − xik − xl k−1 ) i  l ∈ I, k ∈ K 1 (4)
J Set of outbound trucks, |J| = m CiI = S iI + ti i∈I (5)
P Set of products shipped from the inbound to the out- 
bound trucks, |P| = c y jh = 1 j∈J (6)
K Set of sequence positions at the inbound gate, |K| = n h∈H

K1 K \ {1} y jh = 1 h∈H (7)
H Set of sequence positions at the outbound gate, |H| = j∈J
m S Oj ≥ ClO − M(2 − y jh − yl h−1 ) j  l ∈ J, h ∈ H 1 (8)
H1 H \ {1}
ti Processing time of the inbound truck i ∈ I C Oj = S Oj + τj j∈J (9)

τj Processing time of the outbound truck j ∈ J zipj = dip i ∈ I, p ∈ P (10)
MI Sum of the inbound trucks processing times, M I = j∈J
 
i∈I ti
M

Sum of all trucks processing times, M = M I + j∈J τ j zipj = r jp j ∈ J, p ∈ P (11)
i∈I
dip quantity of product p ∈ P delivered by truck i ∈ I
r jp quantity of product p ∈ P required by truck j ∈ J zipj ≤ uipj vipj i ∈ I, j ∈ J, p ∈ P (12)
uipj maximum quantity of product p ∈ P transferable S Oj ≥ +CiI − I
M (1 − vipj ) i ∈ I, j ∈ J, p ∈ P (13)
from i ∈ I to j ∈ J, uipj = min{dip , r jp } Cmax ≥ C Oj j∈J (14)
xik ∈ {0, 1} i ∈ I, k ∈ K (15)
We define the following decision variables: y jh ∈ {0, 1} j ∈ J, h ∈ H (16)
vipj ∈ {0, 1} i ∈ I, j ∈ J, p ∈ P (17)
2
 First Author et al. / Procedia Manufacturing 42 (2020) 475–482 477
Monaco and Sammarra / Procedia Manufacturing 00 (2019) 000–000 3

zipj ≥ 0 i ∈ I, j ∈ J, p ∈ P (18) Defining

S iI ≥ 0 i∈I (19)

S Oj ≥ 0 j∈J (20) ρi = λipj ∀i ∈ I ρ = (ρ1 , . . . , ρn ) (22)
j∈J p∈P

In this model, constraints (2) and (3) are the standard assign- σj = λipj ∀j ∈ J σ = (σ1 , . . . , σm ) (23)
j∈J p∈P
ment constraints. They impose that each inbound truck i is as-
signed to one and only one position k in the sequence of trucks
to be discharged at the inbound door. Constraints (4) and (5) al-
low to compute the completion times for the inbound trucks
I. In particular, constraints (4) ensure that S iI is not smaller  
than ClI , whenever l and i are consecutive in the inbound se- s= σ jτ j − MI λipj (24)
j∈J i∈I j∈J p∈P
quence; otherwise they are redundant. Constraints (6) and (7)
are the same as (2) and (3) for the truck sequence at the out-
bound door. Constraints (8) and (9) play the same role of (4) the optimal value of the Lagrangian Relaxed problem (21) can
and (5), since they are needed to compute the completion times be written in the following form:
of the outbound trucks J. Note that, since M is an upper bound
on Cmax , constraints (8) are trivially satisfied for each pair of
trucks that are not consecutive in the outbound sequence. Equa- I
ZLR (λ) = s + ZLR O
(ρ) + ZLR (σ) + M I ZLR
T
(λ) (25)
tions (10) and (11) are the flow conservation constraints, and
ensure that all the products entering the cross-docking centre
will be correctly delivered to the outbound trucks. Constraints I
where ZLR O
(ρ), ZLR T
(σ) and ZLR (λ) are, respectively, the optimal
(12) logically link z’s and v’s variables, imposing that zipj = 0 if values of the following three subproblems:
product p is not sent from the inbound truck i to the outbound
truck j; conversely, if vipj = 1, then the flow variable zipj cannot 

be greater than uipj . Constraints (13) are the cross-docking con-  I
 ZLR (ρ) =

 min ρiCiI
straints. They impose the correct relation between the start pro- PI(ρ) 
 i∈I (26)

 s.t. (2) − (5), (15), (19)
cessing time of the outbound truck j and the completion time of
the inbound truck i, whenever they are involved in the exchange
of some products, while they are redundant when vipj = 0. Fi-
nally, constraints (14) are the definition of the objective function
(1), and (15) - (20) define the domain of the variables.   


   
 O  O
 Z
 LR (σ) = min C
 max
 − σ C
j j 
PO(σ) 
 (27)
3. The Lagrangian Relaxation of the model 


j∈J
 s.t. (6) − (9), (14), (16), (20)
It is easy to recognize that relations (13) play the role of
coupling constraints, that is complicating constraints in a La-
grangian Relaxation framework. If they are relaxed and dual-
ized in the objective function, the resulting problem will de-  T 
compose in three subproblems. To this aim, let λ be a vector of

 ZLR (λ) = min

 λipj vipj
PT (λ) 
 i∈I j∈J p∈P (28)
Lagrangian multipliers associated to constraints (13), such that 
 s.t. (10) − (12), (17), (18)
λipj ≥ 0, ∀i ∈ I, j ∈ J, p ∈ P, λipj = 0 if uipj = 0. The Lagrangian
relaxed problem is the following:
PI(ρ) is a single machine scheduling problem at the inbound

door: 1|| i wiCi , where wi = ρi and Ci = CiI , i ∈ I (we note, in
  passing, that constraints (4) and (5) can be omitted). Therefore,


   


 p p  its optimal solution is obtained applying the Weighted Shortest
ZLR (λ) = min 
 Cmax + λi j CiI − M I (1 − vi j ) − S Oj 


 i, j,p  (21)
 Processing Time (WSPT) [10].
s.t. (2) − (12), (14) − (20) Similarly for the outbound door, PO(σ) is a single machine
scheduling with a non standard objective function: 1||(Cmax −
 O
j w j C j ), where w j = σ j and C j =C j , j ∈ J. However, it
It is well known that, for each choice of the Lagrangian multi- is possible to prove that, whenever j∈J σ j ≤ 1, the optimal
pliers vector λ ≥ 0, the optimal value of the Lagrangian relaxed schedule can be computed by applying the Weighted Longest
problem provides a lower bound on the optimal value of prob- Processing Time (WLPT) rule [10], while the problem is un-

lem (1)-(20), i.e.: Z ≥ ZLR (λ). bounded from below when j∈J σ j > 1.
3
478 First Author et al. / Procedia Manufacturing 42 (2020) 475–482
Monaco and Sammarra / Procedia Manufacturing 00 (2019) 000–000 4

Finally, subproblem PT (λ) is a multi-commodity transporta- In equation (29), γipj (r) is defined by
tion problem with variable upper bound constraints, that natu-
rally separates into c = |P| single commodity problems, one for
each product p ∈ P. For a fixed p ∈ P, the problem is to move γipj (r) = −S Oj (r) + CiI (r) − M I (1 − vipj (r)) (30)
the product p from the inbound trucks i ∈ I, each with a given
supply dip , to the outbound trucks j ∈ J, each demanding r jp ,
satisfying supply and demand constraints. If a positive amount and t(r) is a step-size defined as
of commodity p is transported from i to j, then it can not exceed
uipj , and a fixed cost λipj has to be paid. The objective function
to minimize is the sum of the fixed costs. Therefore each sin- C ∗ − LB(r)
t(r) = α max
 (31)
gle commodity subproblem is a pure fixed charge transportation γipj (r)2
problem [1, 6], and thus it is hard to be solved to optimality, due i∈I j∈J p∈P
to its combinatorial structure. A heuristic procedure for solving
PT (λ) for a given p ∈ P is described in [1]. It consists in relax-
ing constraints (12), solving a standard transportation problem where α is a smoothing parameter, LB(r) is the lower bound
with unit transportation costs wipj = λipj /uipj , and then setting returned by solving the Lagrangian problem (21) at the r-th it-
vipj = 1 if at least one unit of product p is transported from i to eration, and Cmax ∗
is the best makespan found so far. Observe
j. that γipj (r) is the negative of the slack variable in the (i, j, p)
constraint (13) at the current Lagrangian solution, and corre-
sponds to the (i, j, p) component of a subgradient vector of the
Lagrangian function evaluated at the current solution [5]. From
this point of view, LH is actually a subgradient algorithm for
maximizing ZLR (λ) in problem (21), that is for solving the so
called Lagrangian Dual problem. Therefore, LH returns also
4. Lagrangian heuristics the best lower bound, which allows to measure the quality of
the best computed feasible solution. The pseudocode of LH is
Solving PI(ρ), PO(σ), and PT (λ) for a given set of non neg- reported in Algorithm 3 of Appendix A.

ative Lagrangian multipliers, such that i jp λipj ≤ 1 (see the
discussion on PO(σ) in the previous Section), we get the La-
grangian solutions x, C I , y, C O , z, and v. If C I , C O , and v satisfy 5. Numerical results
constraints (13), they also give a feasible
 solution to the model
(1)-(20), of value C̄max = max j∈J C Oj . 5.1. Description of the test suite
In case C I , C O , and v violate some of the relaxed con-
straints (13), a feasible solution can be obtained computing To test our algorithm, we have considered two sets of in-
ȳ, C̄ O , C̄max by very simple recovering heuristics and setting stances A and B. As for the set A, it consists of 250 instances
x̄ = x, C̄ I = C I , v̄ = v, z̄ = z. In the first recovering heuris- of different dimensions in terms of n and m. For each possible
tics H1, the Lagrangian solution y, C O is transformed in a fea- dimension, 10 instances are considered. The instances of the
sible solution simply by increasing the completion time of each set A have been proposed in [7] for a truck scheduling problem
outbound truck by the same amount, while taking the same se- where the transshipment plan is already known, and therefore
quence provided by y. In the second recovering heuristics H2, for a problem that is substantially different from the one we are
the outbound trucks are first sorted by the earliest starting pro- dealing with. However, these instances can be turned into suit-
cessing times which are compatible with the cross-docking con- able instances fitting the model and the algorithm presented in
straints, and then re-scheduled. The steps of H1 and H2 are de- this paper. Such a transformation is detailed in Appendix B.
tailed in Algorithms 1 and 2 reported in Appendix A. We summarize in Table 1 the dimensions of the instances in A,
Our Lagrangian heuristics LH consists in solving, within an after the procedure of transformation, and we refer the reader
iterative scheme, the Lagrangian sub-problems PI(ρ), PO(σ), to the above cited paper for further details. Here we just remark
and PT (λ), executing the recovering algorithms H1 and H2, that ti , i ∈ I and τ j , j ∈ J are uniformly distributed in the range
and varying the Lagrangian multipliers at each iteration. By this 1 to 10.
way many feasible solutions to the cross-docking problem de-
fined by model (1)-(20) are generated and, at the end, LH re- Table 1. Dimensions of instances in the set A.
turns the best one among them. We use the following rule for |I| |J| |P| Total instances
varying the Lagrangian multipliers from iteration r to iteration
A1 n=5 m ∈ {3, 4, 5, 6, 7} c=m 50
r + 1:
A2 n = 10 m ∈ {6, 8, 10, 12, 14} c=m 50
A3 n = 20 m ∈ {12, 16, 20, 24, 28} c=m 50
A4 n = 40 m ∈ {24, 32, 40, 48, 56} c=m 50
  A5 n = 60 m ∈ {36, 48, 60, 72, 84} c=m 50
λipj (r + 1) = max 0, λipj (r) + t(r)γipj (r) (29)
4
 First Author et al. / Procedia Manufacturing 42 (2020) 475–482 479
Monaco and Sammarra / Procedia Manufacturing 00 (2019) 000–000 5

Since the instances of Set A are intrinsically representative with those returned by Cplex. In particular this comparison is
of a particular and easier case of our truck scheduling prob- done in terms of the quality of the solutions returned by the
lem, we have generated a new set of instances (set B), where two methods and in terms of computation times. We report in
the products arriving by the inbound trucks must be optimally Figures 1 and 2, as average values attained on ten instances of
transferred to the outbound trucks. The set B consists of three the same dimension, the best objective function values (Upper
subsets of 40 instances each, differing in the values of n, m, and Bound - UB), along with the best lower bounds (LB) on the
c as detailed in Table 2. To obtain the test suite B, the values optimal makespan, computed by Cplex and LH. The average
of n, m, c of each row in Table 2 have been fully combined computation times, for each subsets of instances, are summa-
each other. Then, for each possible combination, 10 different rized in Table 3.
instances have been generated by choosing Looking at Figure 1, we observe that Cplex and LH have
basically the same behaviour for 3 ≤ c ≤ 32, as far as the up-
• ti , i ∈ I and τ j , j ∈ J uniformly distributed in the range 1 per bounds are concerned. In particular, for 3 ≤ c ≤ 7, i.e.
to 10; for the A1 instances, the best upper bound and the best lower
• dip , i ∈ I, p ∈ P uniformly distributed in the range 0 to bound values computed by Cplex coincide, meaning that for
1000; these set of instances Cplex always returned the optimal solu-
• r jp , j ∈ J, p ∈ P uniformly distributed in the range 0 to tion. This fact is confirmed by the computation times for the
  
i∈I dip , in such a way that i∈I dip = j∈J r jp , p ∈ P; A1 instances in Table 3. On the same instances LH performs
as well as Cplex in terms of UB values, while the Lagrangian
lower bound is smaller than the best lower bound computed by
Table 2. Dimensions of instances in the set B. Cplex. Starting from the smallest instances in the set A2 , the
quality of the solutions returned by Cplex deteriorates and the
|I| |J| |P| Total instances
number of optimal solutions drastically reduces. Actually, for
B1 n ∈ {10, 15} m ∈ {10, 15} c=3 40 these instances Cplex returns four optimal solutions and no op-
B2 n ∈ {15, 20} m ∈ {15, 20} c=5 40 timal one in all the remaining instances of set A (see also Table
B3 n ∈ {30, 40} m ∈ {30, 40} c=7 40 3 ). For c ≥ 32 our algorithm outperforms Cplex in terms of
both UB and LB values, by an amount that increases as the in-
stance dimension increases. In particular, from Figure 1 it is
5.2. Implementation details evident that the best lower bound returned by Cplex is really
weak and, as a consequence, the corresponding optimality gap
The LH Algorithm has been coded in C++ and Cplex 12.8 results to be very high. On the contrary, the Lagrangian LB re-
has been used for benchmarking purposes. We have ran the ex- turned by LH is significantly higher, so producing lower values
periments on a machine equipped with a 3.1 GHz CPU and of the optimality gap and, therefore, a more accurate estimation
16GB of RAM, giving Cplex one hour of time limit and let- of the solution quality.
ting the LH Algorithm to perform at most 500 iterations. The
initial Lagrangian multipliers have been set as follows Table 3. Average computation time (seconds).

Set Cplex LH

 1 A1 1.53 0.06



 ∀i ∈ I, j ∈ J, p ∈ P : uipj > 0 A2 3483.87 0.38

 nmc
λipj (0) = 
 A3 3600.00 3.83


 A4 3600.00 48.42

0 otherwise A5 3600.00 221.15

so ensuring that their sum does not exceed one. The smoothing
parameter α in equation (31) has been chosen equal to 10−6 . We
finally recall that in solving the Lagrangian Relaxation prob-
lem, the sub-problem PT (λ) can not be solved to optimality. As
disclosed in Section 3, we solve a relaxation of PT (λ) instead,
that is a standard transportation problem. To this aim we have
adopted one of the well known algorithms for determining a
basic feasible solution to a transportation problem, in particular
the greedy one.
5.3. Discussion of the numerical results
In order to evaluate the effectiveness of the proposed La-
grangian heuristics, we compare the results obtained by LH
5
B1 3565.49 0.19
B2 3600.00 0.71
B3 3600.00 6.09
Passing to the results on the set B, we note that (see Figure
2) Cplex exhibits the same behaviour observed on instances of
comparable size in set A. Actually, it is able to find only one op-
timal solution in subset B1 , while in all the other cases it reaches
the imposed time limit returning solutions with very high opti-
mality gaps. Conversely, the behaviour of LH on the instance
set B is not so similar to the one observed for instances of set
A. In particular, on the smallest size instances the UB values
computed by LH are slightly worse than those of Cplex, while
the opposite relation holds for the largest instances of the subset
B3 . The difference between the two UB values, independently
480 First Author et al. / Procedia Manufacturing 42 (2020) 475–482
Monaco and Sammarra / Procedia Manufacturing 00 (2019) 000–000 6

800.00

700.00

600.00

500.00

400.00

300.00

200.00

100.00

0.00
5-3 5-4 5-5 5-6 5-7 10-6 10-8 10-10 10-12 10-14 20-12 20-16 20-20 20-24 20-28 40-24 40-32 40-40 40-48 40-56 60-36 60-48 60-60 60-72 60-84
c=3 c=4 c=5 c=6 c=7 c=6 c=8 c=10 c=12 c=14 c=12 c=-16 c=20 c=24 c=28 c=24 c=32 c=40 c=48 c=56 c=36 c=48 c=60 c=72 c=80
Cplex Best UB Cplex Best LB LH Best UB LH Best LB

Fig. 1. Comparison of numerical results on Set A instances.

350.00

300.00

250.00

200.00

150.00

100.00

50.00

0.00
10-10 10-15 15-10 15-15 15-15 15-20 20-15 20-20 30-30 30-40 40-30 40-40
c=3 c=5 c=7

Cplex Best UB Cplex Best LB LH Best UB LH Best LB

Fig. 2. Comparison of numerical results on Set B instances.

on its sign, is moderate. Of course, the less satisfactory results times (never reaching seven seconds) LH is able to compute
of LH on Set B are due to the poor accuracy we use in solving feasible solutions of certified good quality.
the subproblem PT (λ) (see discussion on this issue in Section 3
and subsection 5.2 ). However this drawback is widely counter-
balanced both by the computation times and the values of the 6. Conclusions
Lagrangian lower bounds. Actually, in very short computation
In this paper we have presented Mixed Integer Linear Pro-
gram and a Lagrangian heuristic algorithm for a truck schedul-
6
 First Author et al. / Procedia Manufacturing 42 (2020) 475–482 481
Monaco and Sammarra / Procedia Manufacturing 00 (2019) 000–000 7

ing problem at a cross-docking terminal. By means of La- Algorithm 3 Lagrangian Heuristics LH

grangian Relaxation of the model, we have been able to de- 1: LB = −∞ /* Lower bound provided by the Lagrangian
compose the problem into three independent subproblems: two problem */
∗ ∗
scheduling problems at the inbound and outbound doors of 2: x∗ , C I , y∗ , C O , z∗ , v∗ , C max∗
= +∞ /* Best solution */
the terminal, and a multicommodity fixed charge transportation I O
3: x̄, C̄ , ȳ, C̄ , z̄, v̄, C̄ max /*Solution returned by H1 and
problem. Including suitable repairing heuristics in a subgradi- H2*/
ent algorithm scheme, we have designed a Lagrangian heuris- 4: x(r), C I (r), y(r), C O (r), z(r), v(r) /* Lagrangian solution
tic procedure and tested it on a wide set of instances. Numeri- at iteration r */
cal results have shown the effectiveness of our approach, both 5: r = 0
in terms of quality of the computed solutions and computation 6: for all i ∈ I, j ∈ N, p ∈ P do
time. 7: λipj (0) = 0 /* initialize the multipliers*/
8: while r ≤ MaxNumberO f Iterations do
9: Compute ρ and σ

10: if j∈J σ j > 1 then
11: r = MaxNumberO f Iterations
12: else
13: Solve PI(ρ), PO(σ), PT (λ), getting x(r), C I (r), y(r),
C O (r), z(r), v(r)
14: Set x̄ = x(r), C̄ I = C I (r), z̄ = z(r), v̄ = v(r)
15: (ȳ, C̄ O , C̄max )=H1(C I (r), y(r), C O (r), v(r))
∗
16: if C̄max < Cmax then
∗ ∗
Appendix A. Pseudocodes of heuristics H1, H2, and LH 17: Cmax = C̄max , x∗ = x̄, C I = C̄ I , y∗ = ȳ, C O , z∗ = z̄,
∗
∗
v = v̄
18: (ȳ, C̄ O , C̄max )=HA2(C I (r), y(r), C O (r), v(r))
∗
19: if C̄max < Cmax then
∗ ∗
20: ∗
Cmax = C̄max , x∗ = x̄, C I = C̄ I , y∗ = ȳ, C O , z∗ = z̄,
∗
v = v̄
Algorithm 1 Recovering Heuristics H1(C I , y, C O , v) 21: Update the Lagrangian multipliers by eq. (29)
1: ∆max = 0 22: r =r+1
∗ ∗
2: for all i ∈ I, j ∈ J, p ∈ P do 23: return x∗ , C I , y∗ , C O , z∗ , v∗ , C max ∗
, LB
3: if vipj = 1 and CiI − C Oj + τ j > ∆max then
4: ∆max = CiI − C Oj + τ j
5: for all j ∈ J do
6: C̄ Oj = C Oj + ∆max
 
7: C̄ max = max j∈J C̄ O j
8: return ȳ = y , C̄ O , C̄ max Appendix B. The instance set A

The set of instances A has been proposed in [7], for a cross-

docking problem with one inbound and one outbound door,
where the transshipment flows of commodities between in-
bound and outbound trucks are fixed. That is, for each outbound
truck j ∈ J the subset I j ⊆ I of inbound trucks sending some
load units to j is known. With reference to our model (1)-(20),
under this setting constraints (10), (11), (12) and the involved
Algorithm 2 Recovering Heuristics H2(C I , y, C O , v) variables z and v are useless, while constraints (13) become
1: for all j ∈ J do
2: ∆min
j = max{CiO | vipj = 1} /* minimum feasible starting
i∈I
p∈P S Oj ≥ CiI ∀ j ∈ J, i ∈ I j (B.1)
processing time for truck j */
3: Sort the outbound trucks j ∈ J in increasing order of the
corresponding ∆min j
4: Schedule the outbound trucks j ∈ J with respect to the It is easy to transform the instances A so as they can be adopted
previous sorting; C̄ O and ȳ accordingly. by our model and algorithm. The transformation procedure pre-
 compute

5: C̄max = max j∈J C̄ j O serves the original dimensions in terms of |I| = n and |J| = m,
while creating suitable values for |P| = c and for dip , r jp , i ∈
6: return ȳ , C̄, C̄max
I, j ∈ J, p ∈ P. It is described in Algorithm 4.
7
482 First Author et al. / Procedia Manufacturing 42 (2020) 475–482
Monaco and Sammarra / Procedia Manufacturing 00 (2019) 000–000 8

Algorithm 4 Instance transformation

1: |P| = |J| = m
2: for all ∈ I, p ∈ P do
3: dip = 0
4: for all j ∈ J, p ∈ P do
5: r jp = 0
6: for all j ∈ J do
7: r j j = |I j |;
8: for all j ∈ J, i ∈ I j do
9: di j = di j + 1

Basically, Algorithm 4 returns an instance for our truck

scheduling problem, starting from an instance defined in [7],
such that:

• the number of products is equal to the number of out-

bound trucks, and each product is required by one and
only one outbound truck.
• each inbound truck carries at most one unit of each prod-
uct.

References

[1] Balinsky, M.L., 1961. Fixed cost transportation problems. Naval Research
Logistics Quarterly 8, 41–54.
[2] Boysen, N., Fliedner, M., 2010. Cross dock scheduling: Classification,
literature review and research agenda. Omega 38, 413–422.
[3] Buijs, P., Vis, I.F., Carlo, H.J., 2014. Synchronization in cross-docking
networks: A research classification and framework. European Journal of
Operational Research 239, 593–608.
[4] Chiarello, A., Gaudioso, M., Sammarra, M., 2018. Truck synchronization
at single door cross-docking terminals. OR Spectrum 40, 395–447.
[5] Fisher, M.L., 1981. The lagrangian relaxation method for solving integer
programming problems. Management Science 27, 1–18.
[6] Fisk, J., McKeown, P., 1979. The pure fixed charge transportation problem.
Naval Research Logistics 26, 631–641.
[7] Fonseca, G.B., Nogueira, T.H., Ravetti, M.G., 2019. A hybrid lagrangian
metaheuristic for the cross-docking flow shop scheduling problem. Euro-
pean Journal of Operational Research 275, 139–154.
[8] Ladier, A.L., Alpan, G., 2016a. Cross-docking operations: Current research
versus industry practice. Omega 62, 145–162.
[9] Ladier, A.L., Alpan, G., 2016b. Robust cross-dock scheduling with time
windows. Computers & Industrial Engineering 99, 16 – 28.
[10] Pinedo, M.L., 2016. Scheduling. Theory, Algorithms, and Systems.
Springer, Cham.
[11] Van Belle, J., Valckenaers, P., Cattrysse, D., 2012. Cross-docking: State of
the art. Omega 40, 827–846.