7th IFAC Symposium on Mechatronic Systems

7th IFAC Symposium

September on MechatronicUniversity,
5-8, 2016. Loughborough Systems UK
7th IFAC
7th IFAC Symposium
Symposium on Mechatronic
on Mechatronic Systems
Systems
September 5-8, 2016. Loughborough University, UK
September
September 5-8,
5-8, 2016.
2016. Loughborough
Available
Loughborough University,
online
University, UK
UK
at www.sciencedirect.com

ScienceDirect
IFAC-PapersOnLine 49-21 (2016) 431438
A
A customer
customer requirement
requirement driven
driven framework
framework
A
A customer
customer
for design requirement
requirement
synthesis - driven
driven
applied to framework
framework
a washing
for
for design
design synthesis
synthesis -
- applied to a washing
for design synthesis - applied to a washing
applied
machine to a washing
machine
machine

machine

K. Berx M. Friedl M. Witters P. Hehenberger
K.
K. Berx
Berx M. Friedl M. Witters P. Hehenberger
K. Berx M. M. Friedl
Friedl M. M. Witters
Witters P. P. Hehenberger
Hehenberger
Flanders Make, Heverlee, Belgium (e-mail:
Flanders Make, Heverlee, Belgium
kristof.berx@flandersmake.be,
Flanders Make, Heverlee,
Flanders Make, Heverlee, Belgium (e-mail:
maarten.witters@flandersmake.be).
Belgium (e-mail:
(e-mail:
kristof.berx@flandersmake.be,
maarten.witters@flandersmake.be).
Linz Center of Mechatronics,
kristof.berx@flandersmake.be,
kristof.berx@flandersmake.be, Linz, Austria (e-mail:
maarten.witters@flandersmake.be).
maarten.witters@flandersmake.be).
Linz Center of Mechatronics,
michael.friedl@jku.at, Linz, Austria (e-mail:
peter.hehenberger@jku.at)
Linz Center
Linz Center of of Mechatronics,
michael.friedl@jku.at, Mechatronics, Linz,
Linz, Austria
Austria (e-mail:
peter.hehenberger@jku.at) (e-mail:
michael.friedl@jku.at, peter.hehenberger@jku.at)
michael.friedl@jku.at, peter.hehenberger@jku.at)
Abstract: In order to maximize the success of a product, the customers wishes need to
Abstract:
be integrated
Abstract: In
In order
into
order the to
to maximize
entire
maximize product the
the success of
design process. a product,The designthe customers
of a mechatronicwishes need
system to
Abstract:
be integrated In into
orderthe to entire
maximize product the success
success
design
of
of aa product,
process. product,
The
the
the customers
design customers
of a
wishes
wishes need
mechatronic need
system
to
to
can integrated
be
be be dividedinto
integrated intoin athe
the 3 entire
different
entire phases,design
product
product requirements
design process.
process. engineering,
The
The design
design architectural
of
of aa mechatronic
mechatronic design and
system
system
can
detailed
can be
be divided
design. Duringin a 3 different
the phases,
requirement requirements
phase, engineering,
customersengineering,
needs are translatedarchitectural design
into engineering and
can
detailedbe divided
divided
design.
in
in a
Duringa 3 3 thedifferent
different phases,
phases,phase,
requirement
requirements
requirements
customers engineering,
needs are
architectural
architectural
translated into
design
design and
engineeringand
characteristics.
detailed
detailed design.
design. In In the
During
During architectural
the requirement
the requirement phase,
phase,
phase,the topology,
customers
customers consisting
needs
needs are of the
translated
are translated number
into and type
engineering
into engineering
characteristics.
of components, In
characteristics. the
as well
the architectural
as some key propertiesphase, the topology,
(size, material) consisting
of theof of the
different number and
components, typeis
characteristics.
of components, In
as the architectural
well architectural phase,
phase, the topology,
the (size,
topology, consisting
consisting the
the number
ofdifferentnumber and
and type
type
chosen.
of
of This phase
components,
components, as
as has aas
well
well as
as
some
major
some
some
key
impact
key
key
properties
on the final
properties
properties (size,
(size,
material)
product
material)
material)
of
of
of
the
performance.
the
the different
different
components,
components,
components,
is
is
is
chosen. This
Since (mechatronic)
chosen. This phase has a
products major needimpact
to meeton the final
an ever product
increasing performance.
number of performance specifica-
chosen.
Since This phase
phase has
(mechatronic) has a
a major
products major need
impact
impact
to
on
on the
meet the
an
final product
product performance.
finalincreasing
ever performance.
number of performance specifica-
tions
Since
Since and constraints,
(mechatronic)
(mechatronic) recently
products
products computational
need
need to meet an
to meet an designdesign
ever synthesis
increasing
ever increasing approaches
number
number of have been developed
of performance
performance specifica-
specifica-
tions
tions and
to support constraints,
design engineers.
and constraints,
constraints, recently
recentlyThese computational
types of methods
computational synthesis approaches
use computational
design synthesis
synthesis approaches power have been
have been developed
and developed
been computer
developed
tions
to and
support design recently
engineers. computational
These types of design
methods use approaches
computational have
power and computer
models
to
to support
supportto search
design
designthe the design space
engineers.
engineers. These
These in an automatic
types
types of
of methods
methods way. However,
use
use computationalas these methods
computational power
power and focus
and entirely
computer
computer
models
on
models to
maximizing
to search
search the the design
functional
design space
space in
in an
performance,
an automatic
the
automatic way.
customers
way. However,
However, actual as
as these
wishes
these methods
and
methodsneeds focus
may
focus entirely
slip to
entirely
models
on to searchthe
maximizing thefunctional
design space in an automatic
performance, the way. However,
customers actual aswishes
these methods
and needs focus
may entirely
slip to
the
on
on background.
maximizing
maximizing the
the functional performance, the customers actual wishes and needs may slip to
functional performance, the customers actual wishes and needs may slip to
the
This background.
article
the background.
background. presents a novel computational design synthesis approach which integrates design
the
This article presents a novel computational design synthesis approach which integrates design
spacearticle
This
This exploration
article presents
presents anda Quality
novel
aQuality Function Deployment
novel computational
computational design during the
design synthesis
synthesis architectural
approach
approach which design phase.
integrates
whichdesign
integrates The
design
design
space
proposed
space exploration
method
exploration and
allows
and to
Quality Function
generate
Function Deployment
architectures
Deployment during
that
during the
maximize
the architectural
customer
architectural phase.
satisfaction.
design phase. The
The
space
proposed exploration
method and
allows Quality
to Functionarchitectures
Deployment that during the architectural design phase. The The
effectiveness
proposed
proposed is illustrated
method
method allows
allows to by generate
to an industrial
generate
generate case study:
architectures
architectures the maximize
that
that design of the
maximize
maximize
customer
mechanical
customer
customer
satisfaction.
architecture
satisfaction.
satisfaction. The
The
effectiveness
for a washingis
effectiveness is illustrated
ismachine.
illustrated by by
by anan industrial
an industrial
industrial case case study:
case study:
study: the the design
the design
design of of the
of the mechanical
the mechanical architecture
mechanical architecture
architecture
effectiveness
for a washing illustrated
machine.
for
for a
a washing
washing machine.
machine.
2016, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: QFD, House of Quality, Formal Modeling, Computational Design Synthesis,
Keywords:
Keywords: QFD,
Architectural QFD, Design House
House of of Quality,
Quality, Formal
Formal Modeling,
Modeling, Computational
Computational Design Design Synthesis,
Synthesis,
Keywords:
Architectural QFD, Design House of Quality, Formal Modeling, Computational Design Synthesis,
Architectural
Architectural Design Design
1. INTRODUCTION generated are determined by the objectives and constraints
1.
1. INTRODUCTION generated are determined
1. INTRODUCTION
INTRODUCTION used in the
generated
generated are determined by
areoptimization.
determined For
by
by
the
thea objectives
the manufacturer
objectives
objectives
and
and constraints
and to gain an
constraints
constraints
used
edge
used in
through
in the
the optimization.
the use
optimization. of For
DSE,
For a
a manufacturer
the optimization
manufacturer to problem
to gain an
gain an
Customer satisfaction is one of the most important aspects used edge in
throughthe optimization.
the use of For athe
DSE, manufacturer
optimization to problem
gain an
Customer
in productsatisfaction
Customer is
is one
design, determining
satisfaction one of of the most
most important
a companies
the economic
important aspects
suc- edge
aspects beingthrough
edge solved needs
through the useusetoof ofbeDSE,directly
DSE, related the customer
the optimization
optimization problem
Customer satisfaction is one of the most important aspects being solved the
needs to be the
directly related problem
the customer
customer
in product
cess. In
in product design,
order
product design, to determining
meet
design, determiningthe a companies
customers
determining aa companies wishes
companies economic economic
in the
economic suc- suc-
best
suc- being needs.
being solved needs to be directly related the
in needs. solved needs to be directly related the customer
cess. In
In order
possible
cess. way, their
order to
to meet the
the customers
demands
meet need to bewishes
customers in
in the
systematically
wishes the bestin- needs.
best needs.
cess.
possible In order to meet the customers wishes in the best According to Chan and Wu (2002), Quality Function De-
tegrated way,
possible
possible way,
their
into the
way, their
their
demands
product
demands
demands designneed
need
to
to be
needprocess.
to be systematically
be This is especially
systematically
systematically
in- According to
in-
in- ployment
According Chan
Chanisand
(QFD)
to andan Wu (2002),
overall
Wu Quality
QualityforFunction
approach
(2002), translating
Function De-
De-
tegrated
true in the
tegrated into
into the product
home-appliance
the product design
product design process.
industry,
design process.where This
This is especially
competition
is especially
especially According
ployment
is ployment to Chan
(QFD) isandan Wu (2002),
overall Quality
approach Function
for translatingDe-
tegrated
true in theinto the
home-appliance process.
industry, where This is
competition is the qualitative
ployment (QFD)
(QFD) customer
is an
is requirements
an overall
overall approach
approach intoforthetranslating
for appropri-
translating
hard,
true in price-levels
in the are
the home-appliance continuously
home-appliance industry, attacked
industry, where while
where competition at
competition is theis the the qualitative
ate quantitative customer
qualitative customer requirements
technicalrequirements
customer requirementsinto
requirements into the
for the
into appropri-
eachappropri-
the stage of
appropri-
true
hard, price-levels are
same time
hard,
hard, price-levels are continuously
new technology
price-levels are keeps driving
continuously
continuously
attacked
attacked
attacked
while
while at
capabilities
while at
the
the the
at for-
the
ate qualitative
quantitative
product
ate development
quantitative technical
technicaland requirements
production.
requirements The for each
results
for each ofstage
suchof
stage ofa
same
ward.
same time
One
time new
example
new technology
is
technologythe keeps
washing
keeps driving
machine,
driving capabilities
a well-known
capabilities for-
for- ate quantitative
product development technicaland requirements
production. The for each of
results stage
such of
aa
same
ward. time
One new technology
example is the keeps driving
washing machine, capabilities
a well-known QFD approach
product
for- product development
developmentcan be used
and
and within
production.
production. the architectural
The
The results
results of
of design
such
such a
home appliance
ward. One example
example which has washing
is the
the existed for
washing severalaadecades.
machine, well-known QFD QFD approach
stage approach
to evaluate
approach can
canand be used
be usedcompare
used within
within the architectural
different architectures
the architectural
architectural design
in
design
ward.
home One
appliance which is has existed machine,
for several well-known
decades. QFD can be within the design
home appliance which has existed for several decades. stage
order to
stage to evaluate
select
evaluatethose and
and thatcompare
maximize
compare different
customer
different architectures
satisfaction.
architectures in
in
home appliance
According which has
to Berliner andexisted
Brimson for (1988),
several design
decades.deci- stage to select
evaluate and compare different architectures in
According to Berliner and Brimson (1988), design deci- order
order to
to select those
those that
that maximize
maximize customer
customer satisfaction.
satisfaction.
sions made to
According
According in Berliner
to the earlyand
Berliner design
and stages (1988),
Brimson
Brimson have a design
(1988), major influ-
design deci- order
deci- In thistowork, selectthethose
authorsthat show
maximize customer satisfaction.
the integration of an exist-
sions madethe in the
the early
early design stages
finalhave aa major influ-
influ- In this work, the authors show the integration of
ence on
sions
sions
ence
made
made
on the
characteristics
in
in the early
characteristics
design
designof the
of
stages
stages
the final
design.
have
have a
design.
An impor-
major
major
An influ-
impor- ing
In
In framework
this
this work,
work, the for
the DSE,
authors
authors as described
show
show the by Berx
integration
the integration etofal.an
ofal.
an
exist-
an(2014),
exist-
exist-
tant
ence stage
on the is the architectural
characteristics of design,
the final in which
design. the
An archi-
impor- ing
with
ing framework
a QFD
framework for DSE,
analysis.
for DSE, as
The
as described
resulting
described by
by Berx
approach
Berx et
et (2014),
allows
al. to
(2014),
ence
tant on
stagethe ischaracteristics of the final design. An impor-
tecture
tant
tant stage
stage is the
of the
is new
the
the
architectural
architectural
architectural
design,
design, in
system is synthesized,
design, in which
in fixingthe
which
which thearchi-
the
the com- with
archi-
archi-
ing framework
aa QFD
accelerate
with QFD
for DSE, The
theanalysis.
architectural
analysis.
as described byapproach
The resulting
design stage,
resulting
Berx et al.
while allows
approach
(2014),
ensuring
allows to
to
tecture of the new system is synthesized, fixing the com- with
com- accelerate a QFD the analysis. The resulting approach allows to
ponents of
tecture
tecture
ponents
ofused
the and
the
used
newtheir
new
and
system
system
their
keyis
key
isproperties.
synthesized,
synthesized,
properties.
Many
Many
fixing
fixingresearchers,
the com-
the
researchers, the integration
accelerate
accelerate the architectural
the of the customer
architectural
architectural
design
design
stage,
stage, while
designrequirements,
stage, while
ensuring
while resulting
ensuring
ensuring
such
ponents as Wyatt
used andet al.
their(2012)
key and
properties.Lin et al.
Many (2009), have
researchers, the
in
the integration
designs
integration that of the
maximize
of the customer
customer
customer requirements,
satisfaction.
requirements, resulting
This ap-
resulting
ponents
such as used and their key properties. Many researchers,
as Wyatt
explored
such design et
Wyatt al.
al. (2012)
etspace exploration
(2012) and
and Lin Lin(DSE)et
et al. (2009),
al.tools
(2009), have
have the
to auto- in
proach
integration
designs is that
applied
of the customer
maximize
to a washingcustomer requirements,
machine satisfaction.
use case.
resulting
This
Section ap-
such
explored as Wyatt
mate thisdesign
et al.
space (2012)
explorationand Lin
novel (DSE)
et al. (2009),
tools
tools to
have
auto- is in
in designs
designsapplied
proach
that maximize
that maximize customer
customer satisfaction.
satisfaction. This ap-22
This ap-
explored
explored
mate this
stage and
design
design
stage
space
space
and
generate
exploration
exploration
generate novel
architectures.
(DSE)
(DSE) tools
architectures.
toThis
to auto-
auto-
This is shortly is
proach
proach is applied to
isdescribes
applied to
aa washing
related
to a washing
machine
research
washing in theuse
machine
machine use
use
case.
case.ofSection
areas
case. washing22
Section
Section
achieved
mate this by formulating
stage and the
generate architectural
novel design
architectures. stage
This as
is shortly
machine
shortly describes
design,
describes related
QFD
related and research
DSE.
research in the
Next,
in the areas
sections
areas of
of washing
3, 4 and
washing
mate thisby
achieved stage and generate novel architectures. This as is
a numerical
achieved
achieved by formulating
by optimizationthe
formulating
formulating the
the
architectural
problem
architectural
architectural
design
design stage
and translating
design stage
stage to shortly
it as
as machine
5
describes
respectively
machine design,
design,
related
QFD
describe
QFD and
and
research
the DSE. in the sections
Next,
application
DSE. Next,
areas of 3,
of QFD
sections
washing
3, 4 and
to
4 and
the
aaannumerical
appropriate
numerical optimization
solver.
optimization problem
However,
problem the and
value
and translating
of the
translating it
designs
it to
to machine
55 respectively design,describe
QFD and the DSE. Next, sections
application of QFD 3, to4 and
the
a
an numerical optimization problem and translating it to respectively describe the application of QFD to the
an appropriate
an appropriate solver.
appropriate solver. However,
solver. However, the
However, the value
the value of
value of the
of the designs
the designs 5 respectively describe the application of QFD to the
designs
Copyright
2405-8963 2016,
2016 IFAC
IFAC (International Federation of Automatic Control)431 Hosting by Elsevier Ltd. All rights reserved.
Copyright 2016 IFAC 431
Copyright
Peer review 2016
under IFAC
responsibility
Copyright 2016 IFAC of International Federation of 431
Automatic
431Control.
10.1016/j.ifacol.2016.10.642
2016 IFAC MECHATRONICS
432 K. UK
September 5-8, 2016. Loughborough University, Berx et al. / IFAC-PapersOnLine 49-21 (2016) 431438
design of a washing machine, the analysis of the design
problem and the resulting optimization problem. Section
6 compares the different generated washing machine archi-
tectures. Finally, section 7 presents some conclusions and
directions for future research.
2. RELATED WORK
2.1 Washing machine design
According to Nygards and Berbyuk (2014), washing ma-
chine research focuses on reducing energy consumption,
increasing capacity and preventing walking (the machine
moving from its position due to the unbalance forces dur-
ing drum rotation).
In recent years, the washing machine is seen more as
a part of a washing and drying system, increasing the
importance of the drying process. The more water that
can be extracted through spinning of the laundry, the
less has to be removed by other means such as heating
which consumes of lot of energy. This causes the spinning
performance to be taken more into consideration when
the energy consumption grading is determined (see Energy
Star (2013)), which makes increasing the spinning speed a
new driving force towards improving this rating. However,
increasing the spinning speed also amplifies the undesired
vibrations and noise, which causes discomfort to the user
and reduces the lifetime and reliability of operation of the
machine.
Increasing capacity is a second focus point in washing ma-
chine research. Customers want to wash a bigger amount of
laundry per cycle. More laundry per cycle requires a bigger
drum diameter and causes a higher amount of (probably
eccentric) spinning mass which in turn results in higher
dynamic load imbalance during operation. This causes
unwanted side-effects, such as vibrations and noise at high
spin revolutions up to walking washing machines.
This walking behavior is an important topic in washing
machine research. The current trend is towards lightweight
plastic and composite components. However, the reduc-
tion of mass associated with these changes increases the
possibility that a washing machine will walk. The topic
of walking is treated with the creation of static design
conditions to be fulfilled in order to avoid this undesired
behavior.
Some authors have studied the impact of some of the
design decisions made during the architectural design
phase on these different aspects. Conrad and Soedel (1995)
presents a simple model for a comparison of the walking
behavior of top- and front-loading machines. In Nygards
and Berbyuk (2014) a multi-body model of the washing
machine is created. This model is used to optimize the
suspension system with respect to vibration output. To
the authors best knowledge, no work exists on a more
complete optimization of the washing machine architec-
ture taking into account all the customer requirements
described above.
2.2 Design Space Exploration
According to Saxena and Karsai (2010), a design space is
the scope of all possible design choices and design space ex-
432
ploration (DSE) is the process of automatically searching
through this design space to find specific design alterna-
tives that satisfy various design constraints and are best
with respect to one or more objective functions. DSE has
been successfully applied within many domains such as
software deployment on a distributed set of ECUs (Denil
(2013)), software product line configuration (Saxena and
Karsai (2010)), electrical circuit design (Koza (2002)) and
mechanical system design (Wyatt et al. (2012)).
DSE techniques can be applied during different stages
of the design. Research areas range from the conceptual
design stage (Helms et al. (2009)), over the architectural
design stage (our research, and others), into the detailed
design stage, where typically only parametric optimization
is applied (Shah et al. (2012)).
These DSE frameworks all contain similar steps: (i) repre-
sent the design space, (ii) evaluate design alternatives and
(iii) explore the design space using guidance algorithms to
find feasible, near-optimal and optimal designs.
When using DSE in the architectural design phase, the
different architectures generated are usually evaluated
and compared through the use of metrics (Wyatt et al.
(2012)). However, since these metrics are the designers
interpretation of an architectures optimality, the resulting
designs do not necessarily maximize customer satisfaction.
2.3 QFD
Quality Function Deployment (QFD) is a matrix-based ap-
proach that aims to consider the customer wishes through-
out the entire product design process. QFD was developed
and first applied in Japan in late 60s. Akao (1972) was the
first article on Quality Deployment to present the ideas,
developments and findings. After ten years of ongoing
development of QFD in Japan, King (1987), Hauser and
Clausing (1988) were the first to learn of QFD in the US
according to Chan and Wu (2002).
One widespread approach to QFD, introduced by the
American Supplier Institue (ASI) (see ASI (2016)) consists
of only four matrices as shown in Figure 1. The matrices
are called Houses of Quality (HoQ). Compared to the
Japanese and Kings approaches which utilizes up to 30
matrices, the ASI-approach is easier to use.
The typical ASI-approach of QFD centers around four
design stages or Houses of Quality:
(1) Product Planning (HoQ-1) is used to translate qual-
itative customer requirements (CRs) to measurable
engineering characteristics (ECs);
Fig. 1. Four phase QFD approach by ASI
 2016 IFAC MECHATRONICS
September 5-8, 2016. Loughborough University,
K. UK
Berx et al. / IFAC-PapersOnLine 49-21 (2016) 431438 433
(2) Parts Deployment (HoQ-2) establishes the relation-
ship between engineering characteristics and various
parts of the product;
(3) Process Planning (HoQ-3) explores the relationship
between parts characteristics and key process opera-
tions;
(4) Production Planning (HoQ-4) relates key process
operations to production requirements.
The results from one HoQ are used as input to the subse-
quent HoQ in order to map customer wishes throughout
the whole product development process.
Figure 2a highlights the basic contents of a HoQ-1. Follow-
ing the guideline of Punz and Hehenberger (2015) the ini-
tial input is the voice of the customer in form of customer
requirements (CRs). Each requirement can be numerically
weighted regarding its importance. The customers percep-
tion regarding competitor products can be analyzed in
the section of market competitive assessment. Competitors
engineering characteristics can be analyzed in the section
of technical competitive assessment. The next input is
the voice of the designer the engineering characteristics
(ECs) together with their direction of improvement; e.g.
maximize (+), minimize (-) or to keep (o) a value of an
EC. Changing an EC in the direction of improvement will
lead to an improved perception of benefit for the customer.
The correlation matrix is located at the top of the HoQ-
1, which allows to model positive or negative correlations
between numerous ECs. The body of the HoQ-1 is called
relationship matrix. It is used to model positive influ-
ences (correlations) of engineering characteristics to the
customer requirements. A common numerical weighting
for the correlations is 9/3/1 representing strong, mod-
erate and weak correlation respectively. This non-linear
scaling of the relation supports to identify ECs that have
the greatest influence on CRs and subsequently for the
product quality. ECs can be quantified by defining target
or limit values in the lower section of the HoQ-1. The result
of the HoQ-1 are the weights of the different ECs. Each
EC weight is calculated as a sum of all weighted CRs times
the associated correlations in the body of the matrix. The
resulting weight of an engineering characteristic quantifies
its influence on customer satisfaction.
3. QUALITY FUNCTION DEPLOYMENT
When applying the presented approach to the case study
of the washing machine, the first step is to gather customer
requirements (CRs) and prioritize them. A QFD HoQ 1 is
used to model the relations between the CRs and the en-
gineering characteristics (ECs). As a result, designers get
a better insight how changes (preferably improvements) of
ECs will contribute to customer satisfaction. Next, it is
demonstrated how customer satisfaction can be included
in the architectural design stage.
Based on a survey eleven CRs were chosen to represent the
customers voice. Since this work focuses on the impact of
the mechanical aspects on the customer satisfaction, the
drum dimensioning, drivetrain properties and the washing
machines vibrational response are used as ECs. The
vibrational response EC is decomposed into three separate
ECs, as proposed by Nygards and Berbyuk (2014):
433
Transferred vibrations: Since the horizontal in plane
vibrations are negligible with respect to the vertical
vibrations, only the force oscillations in the vertical
direction are measured at the four legs of the washing
machine.
Minimal friction coefficient: In modern households, it
is no longer acceptable to fix the machine to the floor.
The washing machine needs to be designed such that no
slip occurs at its legs, even on a slippery surface (e.g.
a rug). This can be achieved by minimizing the friction
coefficient between the machine legs and the floor that is
required to prevent walking during washing and drying.
Bracket movement margin: The bracket, which holds
the motor, drive-train and drum in place, will vibrate
due to the rotating unbalance mass of the clothes in the
drum. The suspension system will suppress this motion,
in order to prevent collisions between the bracket and
the outer shell of the washing machine. The bracket
movement margin is the minimal distance between them
during washing and drying. The larger this margin,
the longer it will take until wear of the springs and
dampers results in collisions between the bracket and
shell, causing noise and possible failure.
As discussed in Section 2.3, usual QFDs only use posi-
tive weights to model correlations within the relationship
matrices. This leads to uncertainty when a single EC cor-
relates with multiple CRs in a positive and negative sense.
Leary et al. (2005) proposes enhancements to QFD by
allowing also negative weights in the relationship matrix.
Now it is possible to trace how improving an EC can both
improve CRs and deteriorate CRs. The resulting HoQ-1 is
shown in Figure 2b.
By analyzing the optimization weight / importance indi-
cators, this HoQ-1 is used by the design teams to improve
customer satisfaction. For example, increasing the drum
volume followed by reducing the vibrations transferred to
the floor while washing and spinning, promises a strong
positive effect on customer satisfaction. To provide the
design teams with a guideline on the expected value of
an EC, a target and a limit value per EC is needed. These
result from previous design experience or products of com-
petitors. Correlations between several ECs are noted in
the top part of the HoQ. By linking the HoQ-1s results to
the subsequent design space exploration and architecture
generation, the derived solutions can be evaluated with
respect to customer satisfaction.
4. DESIGN PROBLEM ANALYSIS
The methodology described in Berx et al. (2014) describes
multiple steps to solve an architectural design problem.
The first step is to formally model the design space in
SysML (Moore and Steiner (2008)) and OCL (Warmer
and Kleppe (2003)) using the profile described by Gadeyne
et al. (2014). This formal model intuitively and graphically
describes the different component choices and the resulting
key properties to be fixed. Creating this model (i) acceler-
ates the process of finding a clear and concise formulation
of constraints and objectives, (ii) allows discussing the
problem with domain experts without exposing them to
how to solve it and (iii) helps in choosing an appropriate
solution algorithm.
2016 IFAC MECHATRONICS
434 K. UK
September 5-8, 2016. Loughborough University, Berx et al. / IFAC-PapersOnLine 49-21 (2016) 431438

Fig. 2. (a) HoQ-1 Overview, (b) HoQ-1 for the washing machine case

434
 2016 IFAC MECHATRONICS
September 5-8, 2016. Loughborough University,
K. UK
Berx et al. / IFAC-PapersOnLine 49-21 (2016) 431438 435
The next sections will describe this formal model and its
usage. Several aspects were added to the work of Gadeyne
et al. (2014):
abstract components whose concrete components
need to be fixed
nested components, possibly only within a concrete
component that needs to be selected
integration of QFD analysis results
4.1 Components & key properties
As mentioned previously, this paper focuses on the me-
chanical aspects of washing machine design. The variables
of the DSE problem being optimized are the system com-
ponents and their key properties:
Drum orientation, location and dimensioning
Drive-train contents
Motor technology, orientation and location
Transmission type and location
Suspension design
Spring stiffness
Damping coefficient
The variables in bold describe the washing machine topol-
ogy. This includes the orientation of drum and motor, since
they determine the possible transmission types.
Figure 3 shows the SysML representation of the resulting
topological design space. The red block at the top shows
that the system-under-design, which is the washing ma-
chine, consists of a drum, bracket, suspension and drive-
train. Each of the orange blocks represents an abstract
component type for which a concrete component needs to
be fixed. For example the drive-train is either a conven-
tional or a direct one. In case of a direct drive-train, the
motor is fixed to be a Permanent Magnet Synchronous
Motor (PMSM). In case of a conventional one, one of
three motors can be chosen and an additional component
is present, namely the transmission. The drum and motor
are treated as cylinders, resulting in their length and radius
as dimensions. The suspension system uses a given number
and location of the springs and dampers, no counterweight
is considered, so only the spring stiffness and damper
coefficients are variable.
4.2 QFD integration
In order to apply the results of the QFD analysis described
previously, in the design space exploration algorithm, the
relevant parts are added to the formal design space model.
The relevant parts of the QFD are the different engineer-
ing characteristics with their direction of improvement,
weight, target and limit values.
The formal model will then describe how each of these ECs
is computed as a function of the different variables of the
DSE problem. As shown in the qualified names of figure
4, these ECs are defined at different levels of complexity:
The filling opening is directly linked to the drum outer
radius.
The price, efficiency and weight of the drive-train is
linked to the contents of the drive-train and are thus
defined as properties of the drive-train.
435
The distance of the drum opening to the front-top
corner of the machine depends on the drum size and
location as well as the machine size. It is defined at
washing machine level.
The kinematic ECs such as vibrational output and
bracket margin depend on the dynamic response of
the washing machine to an unbalance mass in the
drum, rotating at a certain speed. In order to deter-
mine these ECs two scenarios are defined, washing
and drying, which each represent a simulation of a
multi-body model.
In Gadeyne et al. (2014), the formulation of the user
constraints and the objective was based on experience of
the design team. In this work, each of the EC limits is
added to the DSE problem as a constraint. The objective
function will be a weighted sum of relative EC deterio-
ration/improvement, as shown in equation 1, resulting in
architectures that maximize customer satisfaction.
  
ECi,mean ECi
f itness = weighti (1)
i
ECi,limit ECi,mean
4.3 Inputs
The washing machine use case requires a large amount
of inputs. These originate from (i) a re-usable component
library describing the dimensions of the different motor
and transmission types and (ii) user inputs which need to
be set for each separate washing machine design.
The different user inputs needed are determined through
discussions with domain experts on the information needed
to determine the engineering characteristics described in
the previous section. Figure 5 shows all user inputs iden-
tified.
The figure shows that the user inputs can be defined at
the level of analysis scenarios, the system-under-design,
the components and the choice of the component type.
5. SOLUTION STRATEGY
As described in Berx et al. (2014), one effective approach
to solve an architectural design problem is by splitting
it into (i) topology generation, (ii) topology evaluation
through property optimization and (iii) intelligent topol-
ogy selection. The exact solver to use at each level is
to be chosen based on the SysML analysis. In this use
case, because of the computationally expensive multi-body
simulations involved, the topology evaluation is further
subdivided into two subproblems, with the variables to
optimize in each chosen based on the QFD analysis.
5.1 Topology generation
Different topology generation algorithms are possible, de-
pending on the number of variables and complexity of the
constraints. In the washing machine case, the topologi-
cal space is limited with few constraints. Approximately
500 possible topologies exist, of which only 50 fulfill the
topological constraints. Because of this limited number, an
exhaustive search algorithm was implemented in Matlab to
generate all topologies.
2016 IFAC MECHATRONICS
436 K. UK
September 5-8, 2016. Loughborough University, Berx et al. / IFAC-PapersOnLine 49-21 (2016) 431438

Fig. 3. Topological design space of a washing machine drive-train. Orange blocks represent abstract component types
for which a concrete component, green block inheriting from it, needs to be chosen.

Fig. 4. Engineering characteristics with their associated direction of improvement, weight, target and limit values. The
qualified name shows the block where the EC is defined.

Fig. 5. All user inputs required for the washing machine use case. Each input is tagged by the RequiredProperty
stereotype.

436
 2016 IFAC MECHATRONICS
September 5-8, 2016. Loughborough University,
K. UK
Berx et al. / IFAC-PapersOnLine 49-21 (2016) 431438 437

5.2 Topology selection

To limit the number of topologies to evaluate, the ECs

depending on just the topology are computed. The topolo-
gies not achieving the limit value are filtered out while the
remaining topologies are ranked and evaluated in order of
increasing customer satisfaction.

5.3 Topology evaluation through property optimization

A topology is evaluated by optimizing the key properties
of its components. In the washing machine use case, the
property optimization fixes the dimensions and location of
the different components as well as some properties of the
suspension system.
Because of the large number of variables, the property op-
timization is split into two subproblems solved in sequence.
Since the ECs depending on the drum dimensions and
location have a higher importance rating than the vibra-
tional response, the first subproblem will only maximize
these ECs by optimally sizing the drum and fitting it into
the bounding box along with the rest of the components.
The optimization of the suspension system is only per-
formed for those topologies that can fulfill the limits on
the ECs depending on the drum sizing and location. The
properties of the suspension system determine the value
of the ECs depending on the vibrational response of a
washing machine architecture. In order to evaluate the
this response, a rigid multi-body (MB) model of the ar-
chitecture is created in HOTInt (Gerstmayr et al. (2013)).
Since this tool uses text files to store both the model and
the simulation results, the algorithm (written in Matlab)
can generate a text file describing the variable parts of
the washing machine architecture, add this to the MB-
model describing the fixed parts, start a simulation of the
HOTInt tool and read the simulated forces and motions
from the simulation results file.
Each architecture requires two simulations of the MB-
model, representing the washing scenario (low speed, high
unbalance mass) and the drying scenario (high speed, low
unbalance mass) represented in the formal model. The mo-
tions and forces resulting from these simulations are then
used to compute the different engineering characteristics
of these scenarios. The formulation to compute these ECs
is described in detail in the work of Nygards and Berbyuk
(2010).
Since this optimization requires a large amount of simula-
tions of a multi-body model, two assumptions were used
to reduce the number of variables in the suspension opti-
mization. Firstly, since all washing machine architectures
that reached the suspension optimization were symmetric
around the zx-plane through the middle of the drum, the
properties of the springs and dampers were made symmet-
ric around this plane as well. Secondly, the spring lengths
are such that the drum is kept horizontal at a given height
when the drum is empty and not spinning.
5.4 Response Surface Modeling approach
In order to reduce the number of multi-body model eval-
uations needed to find the optimal suspension properties
Fig. 6. Automatically generated multi-body model of the
optimal washing machine architecture in HOTInt
(stiffness and damping), a response surface model (RSM)
(Jones (2001)) based approach is used.
The approach starts by evaluating all ECs that depend
on the multi-body simulation at a limited set of sam-
ple points, generated through a latin hypercube sampling
scheme. Using this data, an RSM, a fast analytical approx-
imation, is fitted for each EC. The resulting surfaces are
then used as a fast approximation of the different ECs in
a gradient-based optimization. The resulting optimum is
then used as a new sample point at which the multi-body
model is evaluated. The process of refitting the RSMs,
gradient-based optimization and MB-model evaluation is
repeated until a convergence criterion is met.
To prevent the gradient-based algorithm from getting
stuck in a local optimum, the objective function used is
the expected improvement (Sasena et al. (2000)), which is
an indication of how much improvement of the customer
satisfaction we expect to achieve if we sample at a given
point. This objective function creates a trade-off between
sample density and actual objective value of the surround-
ing points, causing the algorithm to search more globally.
This RSM-based process reduces the number of multi-
body model evaluations until convergence to only several
100s, greatly reducing the total computational time.
6. RESULTS
The described optimization algorithm was run on a dell
Precision M6800 with a 2.8 GHz i7 core and 16 GB
RAM. The topology generation took only 5 seconds, the
first phase of the property optimization took 1 minute,
processing 50 topologies and the second phase of the
property optimization took 4 days, performing 4000 multi-
body simulations in total. Figure 6 shows the multi-body
model of the optimal architecture, a direct-drive top-
loader. Analysis of the separeate ECs showed that this
architecture achieved the highest drum volume, which was
revealed to be the most important EC, while still achieving
an acceptable vibrational response. Table 1 summarizes
the results of the complete optimization procedure, show-
ing the three distinguishable groups of topologies with the
maximal fitness found for each. This fitness represents a
quantification of the customer satisfaction, computed ac-
cording to equation 1. Within each group, several different
topologies exist with slightly differing fitnesses because of
different vibrational response, but these differences are too
small to draw meaningfull conclusions.

437
2016 IFAC MECHATRONICS
438 K. UK
September 5-8, 2016. Loughborough University, Berx et al. / IFAC-PapersOnLine 49-21 (2016) 431438
Table 1. Maximal achievable fitness for the
different topologies
Drive-train type Top/Front loader Transmission Fitness
Direct Top None 1630
Conventional Top Beavel 1170
Conventional Front Straight 630
7. CONCLUSIONS
A methodology has been presented for the integration of
design space exploration and QFD during the architectural
design phase. Goal was to generate architectures that
maximize the customer satisfaction. The methodology was
applied to the design of the mechanical architecture of a
washing machine.
The methodology starts by performing a House of Quality
I analysis, mapping the customer requirements to engi-
neering characteristics of the system. Then a formal model
of the design space is created in SysML, which helps in
(i) determining the relationship between the engineering
characteristics and the decision variables in the system
architecture and (ii) choosing a suitable solution strategy.
The QFD analysis combined with the mapping of ECs to
decision variables supported selecting the order in which
the variables are optimized in this solution strategy. The
resulting algorithm first generates all possible topologies.
For each of these topologies, first the different components
are dimensioned and placed within the bounding box, after
which the suspension system is designed in order to achieve
an optimal vibrational response of the washing machine.
Since the bottom level uses a Multi-Body model of the
washing machine, Response Surface Modeling is used to
achieve an acceptable computational time.
The presented use case was purely mechanical, however,
the presented approach can be applied to any other mecha-
tronic architectural design problem. In the future, appli-
cation of the proposed systematic approach to further
examples will be considered.
ACKNOWLEDGEMENTS
This work has been partially supported by the Austrian
COMET-K2 programme of the Linz Center of Mecha-
tronics (LCM), and was funded by the Austrian federal
government and the federal state of Upper Austria.
REFERENCES
Akao, Y. (1972). New product development and quality assurance
quality deployment system. Standardization and Quality Control,
25(4), 714.
ASI (2016). Quality function deployment. URL http://www.amsup.
com/quality-function-deployment/.
Berliner, C. and Brimson, J.A. (1988). Cost management for todays
advanced manufacturing: The CAM-I conceptual design. Harvard
Business School Press.
Berx, K., Gadeyne, K., Dhadamus, M., Pipeleers, G., and Pinte,
G. (2014). Model-based gearbox synthesis. Proceedings of the
14th Mechatronics Forum International Conference, Karlstad,
Sweden, 599605.
Chan, L.K. and Wu, M.L. (2002). Quality function deployment:
A literature review. European Journal of Operational Research,
143(3), 463497.
438
Conrad, D. and Soedel, W. (1995). On the problem of oscillatory
walk of automatic washing machines. Journal of Sound and
Vibration, 188(3), 301314.
Denil, J. (2013). Design, verification and deployment of software-
intensive systems: a multi-paradigm modelling approach. Univer-
sity Antwerp.
Energy Star (2013). Clothes washers key product criteria. URL
http://www.energystar.gov/index.cfm?c=clotheswash.pr_
crit_clothes_washers/.
Gadeyne, K., Pinte, G., and Berx, K. (2014). Describing the design
space of mechanical computational design synthesis problems.
Advanced Engineering Informatics, 28(3), 198207.
Gerstmayr, J. et al. (2013). Hotint: A script language based frame-
work for the simulation of multibody dynamics systems. In
ASME 2013 International Design Engineering Technical Confer-
ences and Computers and Information in Engineering Confer-
ence. American Society of Mechanical Engineers.
Hauser, J.R. and Clausing, D. (1988). The house of quality. Harvard
Business Review.
Helms, B., Shea, K., and Hoisl, F. (2009). A framework for compu-
tational design synthesis based on graph-grammars and function-
behavior-structure. In ASME 2009 International Design Engi-
neering Technical Conferences and Computers and Information
in Engineering Conference, 841851. American Society of Me-
chanical Engineers.
Jones, D.R. (2001). A taxonomy of global optimization methods
based on response surfaces. Journal of global optimization, 21(4).
King, R. (1987). Better design in half the time: Implementing quality
function deployment (qfd) in america. GOAL, Lawrence, MA.
Koza, J.R. (2002). Automatic synthesis of both the topology and
numerical parameters for complex structures using genetic pro-
gramming. Springer.
Leary, M., Burvill, C., Weir, J., Shafton, A., et al. (2005). Proposed
quality function deployment improvements. In DS 35: Proceedings
ICED 05, the 15th International Conference on Engineering
Design, Melbourne, Australia, 15.-18.08. 2005.
Lin, Y.s., Shea, K., Johnson, A., Coultate, J., and Pears, J. (2009).
A method and software tool for automated gearbox synthesis. In
ASME 2009 International Design Engineering Technical Confer-
ences and Computers and Information in Engineering Confer-
ence, 111121. American Society of Mechanical Engineers.
Moore, A. and Steiner, R. (2008). A Practical Guide to SysML:
Systems Modeling Language. Morgan Kaufmann.
Nygards, T. and Berbyuk, V. (2010). Pareto optimization of
a washing machine suspension system. In 2nd international
conference on engineering optimization. Lisbon, Portugal.
Nygards, T. and Berbyuk, V. (2014). Optimization of washing
machine kinematics, dynamics, and stability during spinning using
a multistep approach. Optimization and Engineering, 15(2).
Punz, S. and Hehenberger, P. (2015). The house of concepts as
a matrix-based problem-solving method in conceptual design.
International Journal of Product Development, 20(6), 435464.
Sasena, M.J., Papalambros, P.Y., and Goovaerts, P. (2000). Meta-
modeling sampling criteria in a global optimization framework.
In Proceedings of the eighth AIAA/NASA/USAF/ISSMO sym-
posium on multidisciplinary optimization, Long Beach, CA.
Saxena, T. and Karsai, G. (2010). Mde-based approach for gener-
alizing design space exploration. In Model Driven Engineering
Languages and Systems, 4660. Springer.
Shah, A.A., Paredis, C.J., Burkhart, R., and Schaefer, D. (2012).
Combining mathematical programming and sysml for automated
component sizing of hydraulic systems. Journal of Computing and
Information Science in Engineering, 12(4), 041006.
Warmer, J.B. and Kleppe, A.G. (2003). The object constraint
language: getting your models ready for MDA. Addison-Wesley
Professional.
Wyatt, D.F., Wynn, D.C., Jarrett, J.P., and Clarkson, P.J. (2012).
Supporting product architecture design using computational de-
sign synthesis with network structure constraints. Research in
Engineering Design, 23(1), 1752.