You are on page 1of 6

Proceedings,16th IFAC Symposium on

Proceedings,16th
Information ControlIFAC Symposium
Problems on
in Manufacturing
Proceedings,16th
Proceedings,16th
Information ControlIFAC
IFAC Symposium
Symposium
Problems on
on
in Manufacturing
Available online at www.sciencedirect.com
Bergamo, Italy, June 11-13, 2018
Information
Information Control
Proceedings,16th
Control
Bergamo, Italy, JuneProblems
IFAC
Problems
11-13, in
in Manufacturing
Symposium
2018 on
Manufacturing
Bergamo, Italy,
Italy,
Information
Bergamo, JuneProblems
Control
June 11-13, 2018
11-13, 2018
in Manufacturing
Bergamo, Italy, June 11-13, 2018 ScienceDirect
IFAC PapersOnLine 51-11 (2018) 1499–1504
Decision Support System to balance inventory in customer-driven demand
Decision Support System to balance inventory in customer-driven demand
Decision Support System to balance inventory in customer-driven demand
Decision Support System to balance inventory
C. Castiglione*. in
A. Alfieri*
C. Castiglione*. A. Alfieri*
customer-driven demand
E. Pastore*
C. Castiglione*.
Castiglione*. A. Alfieri*
Alfieri*
C. E. Pastore*
 A.
C. Castiglione*.
E. Pastore*
E. 
Pastore* A. Alfieri*
*Politecnico diE. 
Torino,
Pastore* Torino, TO 10129
Italy (Tel: 011-090-7201; e-mail: *Politecnico di Torino,

{claudio.castiglione, Torino, TO 10129 , erica.pastore}@polito.it)
arianna.alfieri
*Politecnico di
*Politecnico di Torino,
Torino, Torino,
Torino, TO 10129
TO 10129
Italy (Tel: 011-090-7201; e-mail: {claudio.castiglione, arianna.alfieri , erica.pastore}@polito.it)
Italy (Tel:
Italy (Tel: 011-090-7201;
011-090-7201; e-mail: *Politecnico
e-mail: di Torino, Torino,
{claudio.castiglione,
{claudio.castiglione, TO 10129 ,, erica.pastore}@polito.it)
arianna.alfieri
arianna.alfieri erica.pastore}@polito.it)
Italy (Tel: 011-090-7201; e-mail: {claudio.castiglione, arianna.alfieri , erica.pastore}@polito.it)
Abstract: A relevant task in the assembling industry is the management of the Available-To-Promise
Abstract:
(ATP) A relevant
quantity task in thecomponents
of the numerous assemblingneeded industry foris the management of the Available-To-Promise
Abstract:
Abstract: A relevant
A relevant task in
task in thethe assembling
assembling industryindustry isFinished
is the
Products (FP)
the management
management of the
of
to reduce
the inventory costs
Available-To-Promise
Available-To-Promise
(ATP)
and quantity
Order-To-Delivery of the numerous
(OTD) components
lead time. This needed
issue for
is Finished
more and Products
more (FP)
important to reduce
as both inventory
the fast launchcosts
Abstract:
(ATP) quantity
(ATP) A relevant
quantity of of the task
the numerous in the
numerous components assembling
components needed industry
needed for is the
for Finished management
Finished Products
Products (FP) of
(FP) tothe Available-To-Promise
to reduce
reduce inventory
inventory costs
costs
and
of newOrder-To-Delivery
productsof and (OTD)
a (OTD)
high varietylead time. This
of products issue is
arefor more
amore
sourceand more
of important
competitive as both the fast launch
(ATP)
and
and quantity
Order-To-Delivery
Order-To-Delivery the numerous
(OTD) components
lead
lead time.
time. This
This needed
issue
issue is
is Finished
more and
and Products
more
more (FP) toadvantage,
important
important reduce
as
as both
both the
the
despite
inventory
fast
fast
they
costs
launch
launch
of new products
increase the complexity and a of high
thevariety
component of products
management. are a Thissourcepaperof competitive
proposes advantage,
a Decision Supportdespite they
System
and
of newOrder-To-Delivery
products
of new products and
and a ofa (OTD)
high
high lead
variety
variety time.
of This
products
of products issue areis amore
are a Thissource
sourceand more
of important
competitive
of competitive as both
advantage,
advantage, the fast
despite
despite launch
they
they
increase
(DSS) tothehelpcomplexity
managers theidentify
to component the management.
sets of FPs able topaper
balance proposes
the a Decision
components Support
ATP System
quantities,
of new
increase
increasetotheproducts
the complexity
complexityand a high
of the variety
component
oftotheidentify
component of products
management.
management. are a source
This paper
Thistopaper of competitive
proposes
proposes a advantage,
Decision
a Decision ATP Support
Supportdespite they
System
System
(DSS)
meanwhile help
reducingmanagers OTD lead the
times. the sets sets of FPs able balance the components quantities,
increase
(DSS) to
(DSS) to helpthehelpcomplexity
managers
managers of the
to component
identify management.
of FPs This
able topaper
balance proposes
the a Decision
components Support
ATP System
quantities,
meanwhile reducing OTD to lead identify
times. the sets of FPs able to balance the components ATP quantities,
(DSS)
© 2018,toIFAC
meanwhile help managers
(International
reducing OTD to identify
Federation
lead times. the Automatic
sets of FPs able to balance the components ATPreserved.
quantities,
meanwhile
Keywords: reducing
Intelligent OTD lead
decision supportof systems
times. inControl) Hosting
manufacturing, by Elsevier
Modelling Ltd.
andAll rights
decision making in
meanwhile
Keywords: reducing
Intelligent OTD lead times.
decision support systems in manufacturing, Modelling and decision making in
complex
Keywords:systems, Intelligent Production
decision planning and control.
support systems
systems Job and activity
in manufacturing,
manufacturing, scheduling,
Modelling Production
and decision
decision activity
making in
Keywords:
complex Intelligent
systems, decision
Production support
planning and in
control. Job and activity Modelling
scheduling, and Production making
activityin
control
Keywords:
complex Intelligent
systems, decision planning
Production support systems
planning and control. in manufacturing,
control. Job andand activity
activity Modelling
scheduling, and decision
Production making
activityin
complex
control systems, Production and Job scheduling, Production activity
complex
control systems, Production planning and control.
control 
Job and activity scheduling, Production activity
control  are fundamental to reduce the OTD lead time while keeping
1. INTRODUCTION 
are
small fundamental
inventory costs to reduce the OTD lead
of components, time while
especially keeping
in thekeeping
case of
1. INTRODUCTION  are
are fundamental
fundamental to
to reduce
reduce the
the OTD
OTD lead lead time
time while
while keeping
1.
1. INTRODUCTION
INTRODUCTION small
highly inventory
variable costs
demand of components,
(e.g., Huang especially
et al (2008) in the case ofa
provided
Offering customizable products is nowadays considered one small are fundamental
small inventory
inventory to reduce
costs
costs of
of the OTD lead
components,
components, time while
especially
especially in
in the
the keeping
case
case of
ofa
Offering customizable 1. INTRODUCTION
productsofiscompetitive
nowadays advantage.
considered The one highly highly
stochastic variable
approach demand (e.g., Huang et al (2008) provided
of the most
Offering important sources
customizable products is nowadays considered one small
highly inventory
variable
variable costswith
demand
demand
MonteHuang
of components,
(e.g.,
(e.g.,
Carloespecially
Huang
simulation).
et
et al
al (2008)
(2008) in provided
the case ofaa
provided
Offering
of customizable
thecustomization
most important productsofiscompetitive
sources nowadays advantage.
considered The one stochastic approach with Monte Carlo simulation).
high level, together with the reduction of The highly variable
the stochastic approach demandwith (e.g., Huang et al (2008) provided a
Offering
of
of the customizable
most important
thecustomization
high most important products
sources
sources
level, togetherofiscompetitive
of nowadays
competitive
with the
considered
advantage.
advantage.
reduction of
one
the stochastic
The These approachesapproach with
lead toMonte
a long Carlo
Monte periodsimulation).
Carlo simulation).
capacity plan for each
product
of life
thecustomizationcycle
most important length
sources (especially
of competitive in some industries
advantage. stochastic
These approach
approaches with Monte
leadonto athe Carlo
longforecast simulation).
period capacity plan for each
high
high
productcustomization
life cycle level,
level,
length together
together
(especially with
with the the reduction
reduction
in some of The
of
industries the
the These component,approaches basedlead values.plan Then, the
characterized
high
productcustomization
life by high
cycle level,
length technological
together
(especially with innovation,
the
in reduction
some such
of
industries as These
the component,approaches based leadonto
to aathelong period
period capacity
longforecast capacity
values. plan for
for each
Then, each
the
product life cycle
characterized by length
high (especially innovation,
technological in some industries such as occurrence
These
component, of
approaches baseda final
leadon to customer
a long
the order
period
forecast triggers
capacity
values. planthefor
Then, actual
each
the
automotive
product life orcycle
characterized byelectronic
length
high industry)
(especially
technological lead in tosome
innovation, an increasing
industries
such as component,
occurrence ofbased on customer
a specific
final the forecast order values.
triggers Then,
the the
actual
characterized
automotive by
or the high
electronic technological
industry) innovation,
lead to (Huang such
an increasing as production
component,
occurrence ofof the
based product.
on customer
aa specific
final the forecast A
order large number
values.
triggers of
Then,
the orders
the
actual
complexity
characterized
automotive on
or by component
high
electronic technological management
industry) leadinnovation, et
such al.,
as occurrence
production ofof the final customer
product. order
A large triggers
number the
of actual
orders
automotive
complexity or the
on electronic
component industry) lead to
management an
an increasing
to (Huang increasing
et al., lead to longofof
occurrence
production queues
the a and,customer
final
specific hence,
product. to order
an
A increase
large triggersof the
number the
of waiting
actual
orders
2005); this research
automotive
complexity or the
on relies on
electronic
component themanagement
industry)fact that,
lead atosingle car model
an increasing
(Huang et production
lead to long ofqueues
the specific
and, hence,product. to anA large
increase number
of theofwaiting
orders
complexity on the component themanagement (Huang et al.,
al., lead time to andlong
then of
thethe production flow Atime.
2005); this research
is composed
complexity
2005); this by the
on
research
relies of
hundreds
component
relies
oncomponents,
on the
fact that,thus
management
fact that,
a single
a several
(Huang
single
car tens
car
model
et
model of production
al., lead to
time andlong ofqueues
then queues
of
specific
the and,
and, hence,product.
hence,
production to
to an
flow an large
increase
increase
time.
number
of theofwaiting
of the orders
waiting
2005);
is this research
composed by hundreds relies of oncomponents,
the fact that,thus a single
severalcar tens
model of time lead to
andlong
then queues
of the and, hence,
production to
flow an increase
time. of the waiting
different versions are produced. Moreover, customers request time and then ofexplicitly
the production starts flowfromtime.
2005);
is
is
different
this research
composed
composed by
by hundreds
versions are
relies of
hundreds oncomponents,
of
produced.
the fact that,thus
components,
Moreover,thus
a single
several
several
customers
car tens
model
tens
request of This
of time
This
research
and then ofexplicitly
research the production starts flow
fromtime.
the consideration that
the hence,
consideration that
short
is delivery
composed
different by
versionslead time
hundreds
are (called
of
produced. Order-To-Delivery
components,
Moreover, thus several
customers lead time,
tens
request of forecasts are never 100% accurate, and, discrepancy
different
short
namely
versions
delivery
OTD). lead are
time
However,
produced.
(called
with
Moreover, customers
Order-To-Delivery
high levels of lead
customization, time, This
request This
forecastsresearch
research
are explicitly
explicitly
never 100% starts
starts from
fromand,
accurate, the consideration
the hence,
consideration
discrepancy that
that
different
short versions
short delivery
delivery lead
lead are
time
time produced.
(called
(called Moreover, customers
Order-To-Delivery
Order-To-Delivery lead
leadrequest
time,
time, forecastsbetween
This predicted
research
are and actual
explicitly
never 100% starts demand
from
accurate, isconsideration
the hence,
and, always present.
discrepancy that
namely
the OTD OTD).
lead time However,
is with
compromised high levels
bylevels
both the of customization,
availability of between forecasts are never
predicted 100%
and accurate,
actual demand and, hence,
isthan
always discrepancy
present.
short
namely delivery
OTD). lead time
However, (called
with Order-To-Delivery
high of lead
customization, time, Starting
forecasts from
are this
never consideration,
100% accurate, rather
and, hence, working
discrepancy on
namely
the
the OTD OTD).
lead time
components
However,
needed
with highbylevels
is compromised
towithproduce the both of customization,
the
finalthe availability
product (the soof between between
Starting
increasing
predicted
predicted
from
the this and
forecast
and actual
actual demand
consideration,
accuracy,
demand
a rather
more
is always
isthan
always
reactive working present.
present.
approach on
namely
the
the OTD
OTD OTD).
lead
lead time However,
time is compromised
is compromised high by
bylevels
both of customization,
availability of between
Starting predicted
from this and actual
consideration, demandrather is always
than working present.
on
the components
called needed
Available-To-Promise to produce - ATP),thebothfinal
and
theproduct
the
availability
(the so
assembling
of Startingincreasing
has already
from
the
been
this consideration,
forecast
implementedaccuracy, by a rather
more
Purvis
than
reactive
et al.
working
approach
(2014), who
on
the
the OTD lead
components
components time is
needed compromised
needed to to produce
produce by
the both
the final
final the availability
product
product (the
(the soof Starting
so increasing from
the this consideration,
forecast accuracy, rather
aaPurvis
more than
reactive working
approach on
called
phase Available-To-Promise
(whose time includes - ATP),
both and the assembling increasing
has already the
been forecast
implementedaccuracy, by more reactive
et al. approach
(2014), who
the components
called
called needed
Available-To-Promise
Available-To-Promise to produce -
- ATP),
ATP),thethefinal
and
and
operations
product
the
the
to be
(the
assembling
assembling so has studied
increasing howthe
already to forecast
been react to accuracy,
implementedthe demand by a variation.
more
Purvis Their
reactive
et al. reactive
approach
(2014), who
phase
performed(whoseand the timewaiting
includes time both
due tothethetheoperations
other to be has
production studiedalready been
howallowed
to react implemented
to the by Purvis
demand et al.Their
variation. (2014), who
reactive
called
phase Available-To-Promise
(whose time includes - both
ATP), andoperations
the assembling approach, by a demand
good by manufacturing flexibility
phase
performed
orders).
(whose
As and timewaiting
the
reducing
includes
the timeboth
lead due tocan
time
thethe operations
other production
increase the
to
to be be studied
firm
has
studiedalready
approach, how been
to
to react
howallowed implemented
react to
to the
by the
a demand
good
Purvis et al.Their
variation.
variation.
manufacturing
(2014),
Their who
reactive
reactive
flexibility
phase
performed
performed(whoseand
and timewaiting
the
the includes
waiting time
time both
due
due to
tothethe
the operations
other
other to be approach,
production
production (assured
studied byallowed
how a lean
to react and
to
by agile
the
a approach),
demand
good variation.
manufacturing leadsTheirtoflexibility
areactive
better
orders).
competitiveAs advantage
reducing the (Jasti lead time2015),
et al., can increase
firms firm approach,
the work (assured byallowed
ofaa the
leandemandby aagile
and good manufacturing
approach), leads toflexibility
a better
performed
orders). Asand
As the waiting
reducing the time
lead due tocan
time othermight
theincrease production
the firm satisfaction
approach, thatapproach),
does not correspond aa to the
orders).
competitive
on increasing
reducing
advantage
the ATP
the
(Jasti
of
lead
their
time
etcomponents,
al., can
2015), increase
firms
thus might
reducing
firm (assured
the work
the (assured
satisfaction byallowed
by a the
of lean
leandemandby aagile
and
and agilegood
that does
manufacturing
approach), not leads
leads to
correspond toflexibility
better
better
to the
orders).
competitiveAs reducing
advantage the
(Jasti leadet time
al., can
2015), increase
firms mightthe firm
work forecasts.
(assured by a lean and agile approach), leads to a better
competitive
on increasingadvantage
unpredictability the ofATPthe (Jasti
of their
OTD
etcomponents,
al., 2015), firms
lead time thus
and
might work
reducing
improving the satisfaction
satisfaction
forecasts. of
of thethe demand
demand that that does
does not not correspond
correspond to to the
the
competitive
on
on increasingadvantage
increasing the
the ofATP
ATPthe (Jasti
of
of their
their etcomponents,
al., 2015), firms
components, thus
thus might work
reducing
reducing the satisfaction
the forecasts. of the demand that does not correspond to the
unpredictability OTD lead time and improving forecasts.
customer
on satisfaction.
increasing
unpredictability
unpredictability the ofATPthe
of the
Despite
of their
OTD the
OTDcomponents,
lead
lead
efficiency
time thus
time ofreducing
and
and
assembling
improving
improving the Differently
forecasts.
Differentlytofrom
from Purvis et al. (2014), a new approach is here
Purvis et discrepancy
al. (2014), abetween new approach is here
customer
processes satisfaction.
is higher andDespite
higher the efficiency
thanks to newof assembling
approaches proposed deal Purviswith the the predicted
unpredictability
customer
customer of
satisfaction.
satisfaction. the OTD
Despite
Despite lead
the time
efficiency
thethanks
efficiency and
of improving
of assembling
assembling Differently
Differently
proposed to from
from
deal Purvis
with the et al.
al. (2014),
(2014), aabetween
et discrepancy new
new approach
approach
the is
is here
here
predicted
processes
(e.g., is
Nicosia higher
et al., and
2017), higher
a high variable to new
demand approaches
requires and actual
Differently demand.
from This
Purvis etapproach
al. (2014), proactively
a new selects
approach sets of
is here
customer
processes
processes satisfaction.
is
is higher
higher and
andDespite
higher
higher the efficiency
thanks
thanks to
to new
newof assembling
approaches
approaches proposed
proposed
and actual to
to deal
deal
demand. with
with the
the
This discrepancy
discrepancy
approach between
between
proactively the
the
selects predicted
predicted
sets of
(e.g., Nicosia
stocks.Nicosia
However, et al., 2017),
increasing a high variable
thethanks
ATP to demand
would requires products to be “pushed to customers”, i.e., it creates sets of
processes
(e.g.,
(e.g., is higher
Nicosia et
et al.,
al., and higher
2017),
2017), a
a high
high variable
variable new cause
demand
demand approaches high and
requires
requires
proposed
and actual
actual
products
todemand.
to
deal with
demand.
be “pushed
theapproach
This
This to
discrepancy
approach
customers”,
betweenselects
proactively
proactively
i.e., it
the predicted
selects
creates sets
sets
sets of
of
of
stocks.
inventory However,
costs. increasing
Hence, a the ATP
trade-off solution wouldis cause high
necessary to products
and to
actualto be
demand. offered to
Thisto the
approachfinal customers
proactively by the
selects setsdealers
(e.g.,
stocks.
stocks. Nicosia
However,
However, et al.,increasing
2017), a high
increasing the
the variable
ATP
ATP demand
would
would cause
cause requires
high
high products
products
products to
to be
be
be “pushed
“pushed
offered toto customers”,
customers”,
the final i.e.,
i.e.,
customers it
it creates
creates
by the sets of
of
dealers
inventory
balance costs. Hence,
inventories and lead a trade-off
times solution is necessary to with specific incentives. The main insight is that past demand
stocks.
inventory
inventory However,
costs.
costs. Hence, increasing
Hence, aa trade-offthe (Busogi
trade-off ATP
solution
solution
et al.,
wouldis 2017).
cause high
is necessary
necessary to products
to products
with
to
to be
specific be “pushed
offered
offered to
incentives. toto
The
customers”,
the
the final
finalinsight
main
i.e.,
customers
customers is
it creates
thatby the
bypast sets of
dealers
the demand
dealers
balance inventories and lead times (Busogi et al., 2017). can be
products used
to be to offered
select atoThe settheof products
final customersthat, giventheademand
bypast set of
dealers
inventory
balance costs. Hence,
balance inventories a trade-off solution is necessary to with specific incentives. main insight is
is that
Usually, the demandand
inventories andof lead
lead
Finishedtimes (Busogi
timesProducts
(Busogi et
et al.,
al.,
(FPs) 2017).
is predicted with
2017). can specific
be
available
with
used
specific
incentives.
to select aexploits
components,
incentives.
The
set of
The
main
main
insight
products
the that,
available
insight is
that past
given
capacity
that past
ademand
setthe
in
demand
of
balance
Usually, inventories
the demand andof lead
FinishedtimesProducts
(Busogi et al., 2017).
(FPs) istopredicted can
can be
be used
available used to
to select
components, select aaexploits
set
set of of products
products
the that, given
given aa in
that, capacity
available set of
setthe
of
with specific
Usually, the forecast methods to obtain, according the Bill best be possible way (from the of capacity saturation standpoint).
Usually,
with the demand
specific demand
forecast
of
of Finished
Finished
methods to
Products
Products
obtain,
(FPs)
(FPs) is
according istopredicted
predicted
the Bill can
available
available used
best possible
to
components,
components, select
way include
aexploits
set
exploits
(from thefinal
products
the
the
capacity
that, capacity
available
available
saturation
given
capacity a in
setthe
in
standpoint).
of
the
of
withMaterial,
Usually, the
specific the aggregated
demand
forecast of Finished
methods demand
to Products
obtain, for each is
(FPs)
according component.
topredicted
the Bill This
available set should
components, exploits theproducts
available that are
capacity highly
in the
with
of specific forecast
Material, methods demand
therequirements
aggregated to obtain,for according to the Bill best
eachquantities
component. best possible
possible way (from
way include the
(from thefinal capacity
capacity saturation
saturation standpoint).
standpoint).
Then,
with
of capacity
specific
Material, forecast
the methods demand
aggregated and production
to obtain, foraccording
each to the Bill
component. are This attractive
best
set for
possible
should customers,
way (from the and products
thisproducts
capacity is made that
saturationpossibleare by highly
standpoint). the
of Material,
Then, capacity therequirements
aggregated demand
and for
production eachquantities
component. are This
This
attractiveset
set should
should
for include
include
customers, andfinal
finalthis products
is made that
that
possibleare
are by highly
highly
the
balanced.
of Material,
Then, This
capacity the balance
aggregated
requirements is the
demand
and starting
for
production point
each for
component.
quantities more
are exploitation
This set for
should of the available
includeand final historical
products demand
that are by data.
highly
Then,
balanced.capacity
This requirements
balance is and starting
the production quantities
point for are attractive
more attractive
exploitation for customers,
customers,
of the and
available this
this is
is made
made
historical possible
possible
demand by the
the
data.
sophisticated
Then,
balanced.capacity
Thisapproaches
requirements
balance (Gansterer,
is and starting 2015).point
production Good forecasts
quantities Moreover,
are exploitation multiple criteria andsuch as profitability, capacity
balanced.
sophisticated Thisapproaches is the
balance (Gansterer, the starting
2015).point Good for for more
more
forecasts attractive
exploitation
Moreover,
for customers,
of
of
multiple the available
thecriteria
available this historical
such
is made
historical
as
possible
demand bydata.
demandcapacity
profitability,
the
data.
balanced.
sophisticated
sophisticated This balance (Gansterer,
approaches
approaches is the starting
(Gansterer, 2015).point
2015). Good for
Good more Moreover,
forecasts
forecasts exploitation of
multiple the available
criteria such historical
as demand
profitability, data.
capacity
Moreover, multiple criteria such as profitability, capacity
sophisticated approaches (Gansterer, 2015). Good forecasts Moreover, multiple criteria such as profitability, capacity
Copyright © 2018 IFAC 1537
2405-8963 © 2018, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©under
2018 responsibility
IFAC 1537Control.
of International Federation of Automatic
Copyright
Copyright ©
© 2018
2018 IFAC
IFAC 1537
1537
10.1016/j.ifacol.2018.08.288
Copyright © 2018 IFAC 1537
IFAC INCOM 2018
1500
Bergamo, Italy, June 11-13, 2018 C. Castiglione et al. / IFAC PapersOnLine 51-11 (2018) 1499–1504

saturation, and so on, can be taken into account while possible to try and provide the dealers with the best set of FPs
constructing these sets of FPs. (in terms of capacity saturation given the available
components for last week, profitability, attractiveness for
Specifically, in this work, a Decision Support System (DSS) customer, and so on) to be sold to customers. This set is
is proposed. Its main characteristics are the simplicity and the composed by products with a high probability to be requested
compatibility with the Enterprise Resource Planning (ERP) by customers, calculated over the past FP demand.
framework, thus making it interesting also from an industrial
point of view. Mhiri et al. (2015) discussed a similar The generation of the orders to influence the demand avoid
approach for the semiconductor industry, estimating the load the negative consequence of the customer free choices. If
for machines and balances the workload. Instead, the DSS customers freely choose, it might happen that, in the short
here proposed is based on a two-step approach with the aim period before the MPS freezes (i.e., when the production
of proactively saturating the remaining capacity of schedules is hardly editable), they choose products containing
components, given their fixed maximum capacity (i.e., the unavailable components. In this case, the product cannot be
ATP of components). The first step, called bundle generator, immediately produced and its ODT lead time becomes longer
selects a set of feasible solutions (i.e., saleable FPs) and than promised. Also, customers can choose to assemble in
assigns them an attractiveness attribute, based on past their final product components with low ATP quantity. In this
demand. The second step consists in an optimization model case, there is a high risk of having unsaturated capacity of the
to choose the best set of FPs to be proposed to the customers production plant and high inventory of the least appealing
and their quantity. components
The DSS was tested on a real case study from an automotive The DSS proposed in this paper tackles the problem of
company and the results show its effectiveness both in terms saturating the available capacity (i.e., the portion of capacity
of balancing ATP quantity of components and reducing the not already in use) of the manufacturing process in the not-
ODT lead time, in addition to its practical applicability. frozen period of the ERP time window, by creating Pre-
Configured Orders (PCO) that are appealing for customers
The remainder of the paper is structured as follows. In section and, hence, have a high probability to be sold. These orders
2 the problem is discussed in details. The bundle generator are incentivized by the dealers to be purchased by customers.
and the optimization procedure are described in sections 3 FPs are all the different versions of a certain product,
and 4, respectively. Section 5 describes the application of the identified by all the possible combinations (bundles) of
approach to the real case study and, finally, section 6 components. The DSS aims to suggest what are the versions
concludes the paper. of the product that should be pushed to customers in each
period, in order to saturate the available capacity taking into
2. PROBLEM DESCRIPTION consideration the attractiveness that bundle of components
The problem addressed in this paper is related to a customer have for customers.
driven system managed by an ERP, which sets the pace of the
manufacturing process. In this system, the customers are 2.1 Pre-Configured orders
allowed to choose which components to assemble on the final
product. The problem deals with a single final product, which To achieve high capacity saturation, the PCOs must use
has several thousands of variants (because of different several low-demand (i.e., less appealing) components
combinations of components). Once customers have chosen, bundled with only a limited number of high-demand (i.e.,
their orders become part of the Master Production Schedule more appealing) ones, thus balancing the total ATP quantity.
(MPS). Specifically, the customer orders are transformed in However, the bundle of components in each PCO must be
production orders and given, at the component level, to the such that the FP can have a good attractiveness for customers.
Material Requirements Planning (MRP) module of the ERP, The PCOs set, in fact, is given to the dealers, who try and
which issues purchasing orders, if needed. The MPS and the push the customers demand towards this set of possible FPs.
MRP are periodically updated, with a rolling horizon For example, as incentive to buy a specific FP belonging to
framework. No update is allowed for the period immediately the PCO set, customers can be offered a fixed and short OTD
before the beginning of the production, as purchasing orders lead time, due to the fact that the order is already inserted into
have already been issued and manufacturing capacity the ERP.
allocated. Usually, the ERP considers a time window of some
weeks. The orders belonging to a given part of this period When the un-frozen planning period becomes frozen, the
cannot be changed (namely, they are frozen). For example, PCOs ordered from the final customers enter in production,
assume to be in the week before the planning horizon (week whereas the unsold PCOs are discarded.
0). The firm cannot make any actions to modify production The attractiveness of a PCO is measured through the
orders in the next weeks since the MRP is frozen. In the last available historical demand of FPs. The historical demand of
week of time window, instead, there are some components a FP is translated into the component demand, and the higher
with available capacity and, hence, new orders can be added the demand, the more appealing the component is and, hence,
to be assembled there. To allocate the available capacity of the more attractive a PCO containing it will be. However,
the last week, it is possible to wait for some orders that two problems must be faced when creating PCOs: 1) there
randomly book this capacity, hoping that some suitable (in are technical incompatibilities among some components that
terms of capacity) products is chosen by customers, or it is

1538
IFAC INCOM 2018
Bergamo, Italy, June 11-13, 2018 C. Castiglione et al. / IFAC PapersOnLine 51-11 (2018) 1499–1504 1501

must be avoided in the creation of the bundle; 2) components The BG also provides the attractiveness measure for each FP
can be Compulsory (C) (product must have them nonetheless that may be produce. A standard Dynamic Programming
customer preference), Available (A) if the customer can approach (Poler et al., 2016) has been used to solve this
request them, or Non-Available (NA) if there is no way to problem. Specifically, we are interested in the combined
add it to a certain product. The availability of components measure of attractiveness that a component has when is
may vary over time, implying, for example, that a component bundled with one or more other components. For each
that is currently A may become C or NA in the next component 𝑠𝑠𝑠𝑠𝑖𝑖 (𝑖𝑖 = 1 … 𝐼𝐼) (where 𝐼𝐼 is the total number of
generation of FPs. Keeping trace of these changes is components), a tree is constructed (Fig. 1). The root of the
fundamental both to correctly evaluate the attractiveness of tree is a specific component (𝑠𝑠𝑠𝑠𝐴𝐴 in Fig. 1), whereas each
the components by historical demand and to create feasible path from the root to a leaf is a possible bundle of
PCOs. components. Thus, the whole set of paths from the root to all
the leaves contains all the possible combinations of
3. BUNDLE GENERATOR components, i.e., all the bundles that contains the root
component. In the example in Fig. 1, bundles are: {𝑠𝑠𝑠𝑠𝐴𝐴 , 𝑠𝑠𝑠𝑠𝐵𝐵 ,
Each PCO can be seen as a bundle of components that will be 𝑠𝑠𝑠𝑠𝐶𝐶 }, {𝑠𝑠𝑠𝑠𝐴𝐴 , 𝑠𝑠𝑠𝑠𝐵𝐵 , 𝑠𝑠𝑠𝑠𝐷𝐷 }, {𝑠𝑠𝑠𝑠𝐴𝐴 , 𝑠𝑠𝑠𝑠𝐶𝐶 , 𝑠𝑠𝑠𝑠𝐷𝐷 }. If a bundle is
in the FP. Hence, creating the PCOs corresponds to infeasible, it is discarded from the final set. The attractiveness
generating the bundles of components. of a bundle is computed as the its past number of sales.
The Bundle Generator (BG) is the first step of the procedure The BG gives as output the set of all the feasible bundles
incorporated in the DSS and has two main tasks: 1) it keeps (i.e., the ones that do not contain any incompatibility) and
trace and updates the information about the compatibility and their attractiveness.
the historical demand of components; 2) it identifies all the
feasible FPs (i.e., all the feasible bundles of components).
About the first task, the BG (i) must be always coherent with
the last available product grid, in terms of technical
incompatibility among components, and (ii) must be able to
treat, in terms of historical demand, all the changes that might
happen during the product life-cycle. Three different cases Fig. 1. Bundle tree for the component 𝑠𝑠𝑠𝑠𝐴𝐴
may be identified.
The component was NA and becomes A or C. In this 4. OPTIMIZATION PROCEDURE
case, if the historical demand is available for this component
Once all the feasible bundles have been generated, the set of
in other FPs, then it is used to evaluate its attractiveness in
bundles to propose to the customers must be individuated,
similar bundle of components, otherwise back forecasting
with their related amount. The choice is made by an
methods might be used to project the current sales backwards
optimization model that is the core of the DSS and it should
in time and obtain an attractiveness evaluation that is
be driven by: (i) the ATP quantity of each component, (ii) the
coherent with the score of the other bundles.
profit, (iii) the marginal attractiveness of bundles.
The component was A and becomes C or NA. The
While the ATP quantity is quite straightforward to consider,
two situations must be treated in a different way. If the
it is more difficult to define the profit in a proper way. In fact,
accessory becomes C, it starts appearing in every product
the profit can be interpreted as the sum of the economic
sold. Thus, no specific adjustment to historical sales is
margins of the single components of the FP, or the margin of
required. Instead, if the component becomes NA, it still
the whole FP (not considering it as a mere sum of all the
appears in the historical sales. To avoid generating infeasible
margins of the components). These different interpretations
bundles, all the bundles containing this component should be
lead to different optimization models with different sets of
eliminated from the sale database. However, dropping the
PCOs as solution. In particular, considering the single
historical demand of the related FP would imply losing
economic margins would imply imposing a minimum
historical information about the other components belonging
profitability on each produced bundle, while considering the
to the FP. Hence, instead of dropping the historical orders
aggregate margin might allow the production of
away, they are changed by substituting the NA component
configurations with negative margin if this can increase the
with an A that could have been chosen by the customers
capacity saturation.
instead of it (at the time it was still A). To give an example,
let assume that customers can choose among 4 different The marginal attractiveness of a set of products, instead, is
wheels when buying a specific car. Hence, the 4 wheels are related to the idea of marginal utility developed in
all A components. If one of them becomes NA, then the microeconomics (Kauder, 2015). It assumes that the appeal
historical demand of the car containing the no longer of a set of products should decrease with an increase in its
available wheel is redistributed on the historical demands of production quantities, since its demand has already been
the other 3 wheels, according to their market share. The partially satisfied by products already produced.
component was C and becomes A or NA. This situation is
similar to the one described above and, hence, the same Considering all these aspects in a single optimization model
approach is used. would lead to a multi-objective optimization problem.

1539
IFAC INCOM 2018
1502
Bergamo, Italy, June 11-13, 2018 C. Castiglione et al. / IFAC PapersOnLine 51-11 (2018) 1499–1504

However, as the DSS aims at providing the manufacturing helps reducing the set of generated bundles to be included in
planners with a tool that allows to align in the most accurate the model), the model is quite simple and can be solved by a
way the capacity planning with the future customers demand, commercially available software.
only the capacity plan standpoint is considered. Hence, the
profits are left out from the model, leaving to the sales and to 5. EMPIRICAL RESULTS
the marketing departments the task to commercialize in the
best manner the proposed configuration. Furthermore, DSS The DSS proposed in the previous sections (i.e., the bundle
works with the available capacity left in the last period, generator and the optimization procedure) has been applied to
which usually it is booked in a completely customer-driven a real case study of an automotive company (referred to as
way without optimizing capacity saturation. The DSS Company) that, for reason of confidentiality, will remain
provides an optimal solution of PCOs which, in case they are anonymous. It takes into consideration the Italian market of a
all sold, leads to a good result, whereas, if some of them are unique model year in all its possible combinations, which are
not sold, there is no cost for the Company (as no investment characterized by fuel type (i.e., gasoline or diesel),
has been made at the beginning). displacement, traction, transmission and trimming plus each
eventual optional component.
Let the index 𝑖𝑖 indicate the bundles of components, also
named as FPs (𝑖𝑖 = 1,..., I with I the number of bundles found There is a fixed period, measured in weeks, between the
by the BG), and the index 𝑗𝑗 indicate the components (𝑗𝑗 =1, ..., application of the optimization procedure and the period
J with J the total number of considered components). The (week) where PCOs are planned for. This is related to two
variable 𝑥𝑥𝑖𝑖 represents the quantity of bundle 𝑖𝑖, whereas main factors: 1) the freezing of the production plan in the
variable 𝑔𝑔𝑗𝑗 indicates the unused ATP quantity for component next closer weeks and 2) the complete saturation of some
𝑗𝑗. The attractiveness of bundle 𝑖𝑖 is known from the BG
production constraints during the next following weeks.
solution and represented by the parameter 𝑣𝑣𝑖𝑖 , whereas 𝑎𝑎𝑖𝑖𝑖𝑖 is a The optimization procedure has been designed to be applied 5
parameter equal to 1 if component j is in the configuration of weeks earlier than the beginning of the ERP frozen period so
bundle 𝑖𝑖, 0 otherwise; finally, 𝑐𝑐𝑗𝑗 is the ATP quantity for as to be able to fulfil the unsaturated capacities of specific
component j. The optimization model is the following: components. The ERP of the Company works with a six-
week time horizon. The first five weeks are frozen (i.e., the
max ∑𝐼𝐼𝑖𝑖=1 𝑣𝑣𝑖𝑖 𝑥𝑥𝑖𝑖 − ∑𝐽𝐽𝑗𝑗=1 𝑔𝑔𝑗𝑗 (1)
production schedule cannot be changed), whereas the last
s.t. week is still subject to changes. The DSS is run at the
𝐼𝐼 beginning of the first week of the time horizon and the set of
PCOs are given to the dealers, which push the customers to
∑ 𝑎𝑎𝑖𝑖𝑖𝑖 𝑥𝑥𝑖𝑖 ≤ 𝑐𝑐𝑗𝑗 ∀𝑗𝑗 (2) buy the products contained in the set for weeks before the
𝑖𝑖=1
ERP frozen. At the end of this period, all the PCOs ordered
𝑔𝑔𝑗𝑗 = 𝑐𝑐𝑗𝑗 − ∑𝐼𝐼𝑖𝑖=1 𝑎𝑎𝑖𝑖𝑖𝑖 𝑥𝑥𝑖𝑖 ∀𝑗𝑗 (3) by the customers are added to the production schedule for the
𝑐𝑐𝑗𝑗 , 𝑔𝑔𝑗𝑗 , 𝑥𝑥𝑖𝑖 , 𝑣𝑣𝑖𝑖 𝜖𝜖 ℕ ∀𝑗𝑗, 𝑖𝑖 (4) last week of time window while all the PCOs not sold to any
𝑎𝑎𝑖𝑖𝑖𝑖 𝜖𝜖[0,1] ∀𝑗𝑗, 𝑖𝑖 (5) customers are discarded. The DSS is run at the beginning of
each week in a rolling horizon way. Fig. 2 shows rolling time
Equation (1) is the objective function. The first term refers to window.
the total bundles attractiveness while the second is the
remaining ATP quantity. The objective function represents
the attractiveness maximization together with the
minimization of the unused ATP quantity. Constraints (2) Fig. 2. Six periods of rolling time window
are the capacity constraints on the ATP quantity of each
component, while equations (3) define the unused capacity Each time the DSS is run, the BG creates the most attractive
for each component. The optimization problem maximizes bundles from historical data (section 3), and then the
the attractiveness for the customers, providing a solution with optimization model is instantiated (section 4) and solved by
the largest possible number of products, due to the fact that using the commercially available software ILOG CPLEX
each FP has its attractiveness. The more FPs are considered, version 12.6.1.
the more reduced the available capacity 𝑐𝑐𝑗𝑗 is. The
In this case study, the DSS has been tested on four
optimization does not aim at saturating completely the consecutive weeks. Each week has its own set of constraints,
available capacity, rather at providing a solution which different from the other weeks in the amount of ATP quantity
maximizes the number of products sellable according to their of the several components.
attractiveness. This is the reason why there is no
normalization between the two parts of equation (1). The The performance of the DSS is measured as the effectiveness
attractiveness term, made by sold units in the past demand, is of the production mix suggested by the proposed approach.
of the order of thousands, while 𝑔𝑔𝑗𝑗 is of the order of tens. To evaluate the effectiveness, for each of the four weeks, the
However, because of the presence of several solution with PCOs are compared with the real sales occurred in the period
same attractiveness, among them it is possible to choose the between the launch of the order creation procedure and the
ones with larger capacity saturation. Thanks to the BG (that last day of the frozen period (i.e., the six weeks after the
already evaluates all the configurations sold in the past and launch of the DSS).

1540
IFAC INCOM 2018
Bergamo, Italy, June 11-13, 2018 C. Castiglione et al. / IFAC PapersOnLine 51-11 (2018) 1499–1504 1503

5.1. Components characteristics

The case study deals with approximately 30 components (it


depends on the specific week) composing the products. These
components can be clustered in four families: single
components, mechanical components, wheel components and
radio device components. The first family (single
components) groups components that are not directly linked
one to the other; they usually are top of the line components
as leather seats, refined external paints, keyless device, pilot-
support systems and so on. The second family (mechanical
components) includes several combinations of fuel type,
transmission, traction and displacement. The last two
families contain homogeneous components, one contains Fig. 3. Bar chart for week 1
wheels of each material, while the other groups the two kind
of radio devices integrated with other technologies. The
bundles generated in the first step of the procedure are, on
average, 12000.

5.2. Results

Results are expressed in percentage saturation found in week


X for component j, with X being one of the four weeks in
which the DSS is launched (week 1-4) and j one of the
components considered by the DSS. In each week, the
saturation is always found for the sixth week of the ERP
planning horizon and the four cases refer to the planning
horizon starting in week 1, 2, 3 and 4, respectively. The
percentage is expressed as sold out portion of the ATP Fig. 4: Bar chart for week 2
quantity available at the beginning of optimization process.
Figures 3 to 6 report, for each sixth considered week, the
percentage of saturation for each of the four families of
components. In each figure, the blue bars represent the real
sales of the components occurred in the specific week
(expressed as a percentage of the ATP quantity 𝑐𝑐𝑐𝑐 at the
beginning of that week). The orange bars, instead, represent
the saturation reached by the DSS (again expressed in
percentage). Specifically, if a PCO is sold to a customer, then
its components belong to the orange bar.
The graphs highlight two main results. First, comparing the
real sales (blue bars) and the suggested production mix
(orange bars), for each family, it appears that the suggested Fig. 5: Bar chart for week 3
production mix avoids high demand peaks for specific
components and lead to a more uniform use of the
components of the same family. In fact, for each family, the
height of the orange bars shows a smaller variability than the
one of the blue bars. Hence, a more homogeneous use of the
ATP quantity of components belonging to a family could be
reached by using the DSS.
The second result comes from the difference between the
maximum percentage of ATP sold (100% on the y-axis) and
the current ATP used (height of the bars). This difference
represents the remaining ATP and strictly depends on the
initial inventory of the components; hence, a too big
remaining ATP can be due to a mismatch between actual and
predicted sales (i.e., the mismatch between what customers
Fig. 6: Bar chart for week 4
want and what was predicted in the mid-term planning).

1541
IFAC INCOM 2018
1504
Bergamo, Italy, June 11-13, 2018 C. Castiglione et al. / IFAC PapersOnLine 51-11 (2018) 1499–1504

The fact that the proposed production mix is rather different profitability of a solution have not been taken into account.
form real sales also shows that there is a high amount of Hence, to reach a production mix that is also feasible from a
available capacity that might be allocated in a different way profitability perspective, additional information about the
(avoiding fixed cost, or, when it represents sunk cost, it might economics of the produced components and sold products
be used to reduce losses). Under this perspective, the have to be included in both the procedure phases, together
proposed DSS can be useful in assessing the mid-term with the information about the cost of extra capacity and
forecast for each component made by the Company. unused capacity, since they surely impact on the appealing of
the production mix (both in terms of product configurations
The results discussed above were considered very interesting and produced quantities). The inclusion of such information
by the Company. However, the model here proposed suffers will lead to a more complex multi-objective optimization,
from a quite important limitation from the industrial which is the subject of the on-going research. Future
standpoint. The capacity saturation, as previously mentioned, research must also take into account the profitability of each
should also take into account the economics of the Company. single component integrated within its historical demand
In the current solution procedure (DSS), however, the data, in order to understand its trend and analyse the
economics has not been considered yet. correlations with other components, thus leading to a new
Moreover, in real sales data, demand has never been lost, type of bundle generation.
meaning that when the requested capacity is larger than the
available capacity, the demand is satisfied with extra orders REFERENCES
that might be made through extra costs. This situation has Busogi, M., Ransikarbum, K., Oh, Y. G., & Kim, N. (2017).
been represented in the bar charts with completely satisfied Computational modelling of manufacturing choice
capacity for the given components (i.e., a 100% saturation), complexity in a mixed-model assembly line.
but without considering the cost of the extra capacity. International Journal of Production Research, 55(20),
5976-5990.
6. CONCLUSIONS Gansterer, M. (2015). Aggregate planning and forecasting in
make-to-order production systems. International Journal
This paper deals with the problem of saturating the capacity
of Production Economics, 170, 521-528.
(in terms of ATP quantity of components) of a plant
Huang, George Q., Timothy W. Simpson, and B. Joseph Pine
producing highly customizable products, when the total plant
II (2005). The power of product platforms in mass
capacity is not modifiable as in a short-term period (and,
customisation. International Journal of Mass
hence, only its allocation to the different type of products can
Customisation, 1, 1-13
be chosen). A second issue addressed by this research is the
Huang, M. G., Chang, P. L., & Chou, Y. C. (2008). Demand
effort to match the real demand that is occurring with the
forecasting and smoothing capacity planning for
forecasted components consumption in mid-term.
products with high random demand volatility.
A DSS is proposed whose aim is to saturate the capacity International Journal of Production Research, 46(12),
meanwhile increasing the customer satisfaction. This aim can 3223-3239.
be achieved by offering to the customers a FP that is very Jasti, N. V. K., & Kodali, R. (2015). Lean production:
similar to the desired/requested one together with a smaller literature review and trends. International Journal of
ODT lead time. Production Research, 53(3), 867-885.
Kauder, E. (2015). History of marginal utility theory.
The DSS is composed by two phases. In the first phase, using Princeton University Press.
the information about past sales, the BG generates product Mhiri, E., Jacomino, M., Mangione, F., Vialletelle, P., &
configurations that can be attractive for the customers. In the Lepelletier, G. (2015). Finite capacity planning
second phase, using an optimization model, the best amount algorithm for semiconductor industry considering lots
of each FP found by the BG is computed by considering the priority. IFAC-PapersOnLine, 48(3), 1598-1603.
ATP quantity related to each component belonging to it. Nicosia, G., Pacifici, A. (2017) Scheduling assembly tasks
with caterpillar precedence constraints on dedicated
The proposed approach has been tested on a real case study machines, International Journal of Production Research,
from the automotive sector and the results showed that it can 55(6), 1680-1691.
lead to a more balanced saturation of the capacity for the Poler, R., Mula, J., & Díaz-Madroñero, M. (2016).
same family of components, thus handling in a better way the Operations research problems. Springer London Limited,
fixed production capacity. 325-329.
Although this procedure has not the specific purpose of Purvis, L., Gosling, J., & Naim, M. M. (2014). The
measuring and evaluating the efficiency of the allocated development of a lean, agile and leagile supply network
capacity for each component, it can give useful managerial taxonomy based on differing types of flexibility.
insights. In fact, through the periodic comparison of the International Journal of Production Economics, 151,
components saturation percentage, it is being possible to 100-111.
evaluate a reallocation of capacity in order to increase the
profitability. Moreover, it helps identifying the bottleneck
component. In the current DSS, the economics related to the

1542

You might also like