PrimaVera: Synergising Predictive Maintenance

applied
sciences
A icle
P imaVe a: Syne gising P edic i e Main enance
B am Ton 1, Rob Bas en 2, John Bol e 3, Jan B aaksma 4, Alessand o Di Bucchianico 5,
Philippe an de Calseyde 2, F ank G oo eman 6, Tom Heskes 7, Nils Jansen 7,
Wou e Teeuw 1, Tiedo Tinga 8and Ma iëlle S oelinga 7,9,*
1Saxion Uni e si y o Applied Sciences, 7513 AB Enschede, The Ne he lands; [email p o ec ed] (B.T.);
w[email p o ec ed] (W.T.)
2Depa men o Indus ial Enginee ing & Inno a ion Sciences, Eindho en Uni e si y o Technology,
5612 AZ Eindho en, The Ne he lands; [email p o ec ed] (R.B.); p.p. .m. [email p o ec ed] (P. .d.C.)
3The Hague Uni e si y o Applied Sciences, 2521 EN Den Haag, The Ne he lands; [email p o ec ed]
4Depa men o Design, Uni e si y o Twen e, P oduc ion and Managemen ,
7522 NB Enschede, The Ne he lands; [email p o ec ed]
5Depa men o Ma hema ics and Compu e Science, Eindho en Uni e si y o Technology,
5612 AZ Eindho en, The Ne he lands; [email p o ec ed]
6Royal Ne he lands Ae ospace Cen e, 1059 CM Ams e dam, The Ne he lands; ank.g oo eman@nl .nl
7Ins i u e o Compu ing and In o ma ion Sciences, Radboud Uni e si y,
6525 XZ Nijmegen, The Ne he lands; [email p o ec ed] (T.H.); [email p o ec ed] (N.J.)
8Depa men o Mechanics o Solids, Su aces & Sys ems, Uni e si y o Twen e,
7522 NB Enschede, The Ne he lands; [email p o ec ed]
9Fo mal Me hods and Tools, Uni e si y o Twen e, 7522 NB Enschede, The Ne he lands
*Co espondence: [email p o ec ed] o [email p o ec ed]
Recei ed: 31 July 2020; Accep ed: 19 No embe 2020; Published: 24 No embe 2020


Abs ac :
The ull po en ial o p edic i e main enance has no ye been u ilised. Cu en solu ions
ocus on indi idual s eps o he p edic i e main enance cycle and only wo k o e y speci ic se ings.
The o e a ching challenge o p edic i e main enance is o le e age hese indi idual building blocks
o ob ain a amewo k ha suppo s op imal main enance and asse managemen . The P imaVe a
p ojec has iden i ied ou obs acles o ackle in o de o u ilise p edic i e main enance a i s ull
po en ial: lack o o ches a ion and au oma ion o he p edic i e main enance wo k low, inaccu a e o
incomple e da a and he ole o human and o ganisa ional ac o s in da a-d i en decision suppo
ools. Fu he mo e, an in ui i e gene ic applicable p edic i e main enance p ocess model is p esen ed
in his pape o p o ide a s uc u ed way o deploying p edic i e main enance solu ions.
Keywo ds: p edic i e main enance; p ocess model; in e disciplina y esea ch; case s udies
1. In oduc ion
P edic i e main enance is he abili y o use da a-d i en analy ics o op imise he upkeep o
capi al equipmen [
1
]. P edic i e main enance b idges he gap be ween condi ion-based main enance
and co ec i e main enance and is enabled by he ad en o Indus y 4.0 [
2
]. Value is c ea ed by
ans o ming he collec ed da a om in elligen sys ems in o p edic ions abou he sys em’s heal h,
so ha main enance can be done exac ly when and whe e needed. Es ima es o he impac o p edic i e
main enance a y widely, bu in gene al he e u n o in es men is deemed o be a ou able [
3
].
Despi e he a ou able e u n on in es men , implemen a ion o p edic i e main enance in p ac ice is
s ill limi ed in many indus ies [4,5].
Fu he mo e, p edic i e main enance is a key enabling echnology o se i isa ion in sma
indus ies. Se i isa ion is an eme ging end [
6
] in which o ganisa ions and ci izens no longe own
hei asse s, bu a he lease hei se ices: companies buy hou s on p oduc ion machine ies wi h a
Appl. Sci. 2020,10, 8348; doi:10.3390/app10238348 www.mdpi.com/jou nal/applsci
Appl. Sci. 2020,10, 8348 2 o 19
gua an eed h oughpu ; people lease a ca a he han buying one. As a consequence, se i isa ion
manda es cons an a ailabili y a a low cos , p esc ip i e (pe sonalised) se ice, and ull digi isa ion
and au oma ion o se ice p o ision.
Al hough he p ospec i e bene i s o p edic i e main enance a e emendous, ealising he
en isioned bene i s is a om i ial. While many co e building blocks o p edic i e main enance
(such as senso echnology, ailu e p edic ion me hods, and op imisa ion echniques) exis ,
cu en solu ions ocus on indi idual s eps in he p edic i e main enance cycle, and only wo k
o e y speci ic se ings. The o e a ching challenge o p edic i e main enance is o le e age hese
indi idual building blocks in o an e ec i e and e icien amewo k ha suppo s op imal main enance
and asse managemen in a complex a ena. The P imaVe a p ojec picks up his challenge h ough a
mul idisciplina y eam p o iding exac ly hese expe ises.
This pape highligh s wo majo elemen s o he p ojec o syne gise p edic i e main enance.
Fi s , he challenges hinde ing he success ul applica ion o p edic i e main enance ha e been
iden i ied and a e p esen ed in his pape . Secondly, his pape in oduces a gene ic p edic i e
main enance p ocess model which p o ides a s uc u ed app oach o deploying new p edic i e
main enance solu ions.
The P imaVe a p ojec includes leading indus ial pa ne s om h ee majo sec o s o he Du ch
economy: in as uc u e, high- ech and ma i ime. The p ojec has been awa ded a g an o i e million
eu os in unding om he Du ch Resea ch Council (NWO) and co- unding om he pa icipa ing
conso ium membe s. The p ojec has a du a ion o i e yea s.
The es o his pape is o ganised as ollows. Sec ion 2p esen s he s a e o he a o each o he
six elemen s comp omising p edic i e main enance. Based on he s a e o he a , he P imaVe a p ojec
has iden i ied ou obs acles o o e come in o de o u ilise p edic i e main enance o i s ull po en ial.
These obs acles a e desc ibed in Sec ion 3, Sec ion 4ou lines he gene ic p ocess model o ackle hese
obs acles. Fu he mo e, Sec ion 4desc ibes he esea ch app oach aken o each s ep o he p ocess
model. Sec ion 5de ails he me hodology used du ing he p ojec oge he wi h a b ie o e iew o
he in ended demons a o s. Sec ion 6de ails he cons i uen s o he conso ium and inally, he las
sec ion con ains he conclusion.
2. S a e o he A
P edic i e main enance en ails six s eps; da a acquisi ion, da a p ocessing and diagnos ics,
p ognos ics, op imisa ion o main enance and logis ics, asse managemen , and human and
o ganisa ional ac o s. This sec ion will ou line he s a e o he a o each s ep. La e on asse
managemen and human and o ganisa ional ac o s a e conside ed as one s ep as hese wo s eps a e
closely ela ed.
2.1. Da a Acquisi ion
P edic i e main enance o condi ion moni o ing ha goes beyond he isual inspec ions by a
human inspec o is always da a-d i en. The mos udimen a y o m o da a-d i en main enance
would be he analyses o log iles and e o messages [
7
]. A nex le el is using speci ic sensing
me hods, e.g., ib a ional equency measu emen s a bea ings [
8
], o assess he heal h o componen s.
Mo e ad anced me hods include eal- ime moni o ing which can aise an ala m based on p ede ined
c i e ia. Taking i u he , machine lea ning and big da a analysis o senso da a a e being esea ched [
9
].
In p ac ice, many companies s uggle o inco po a e da a-d i en wo k lows wi hin hei
company [
4
,
5
]. Fi s o all, he e may be no da a, incomple e da a, e oneous da a, unaligned da a o
simply no enough da a. Logging o e o s may be incomple e o da a a e no s o ed a all. To apply
he powe o machine lea ning echniques o iden i y pa e ns, la ge amoun s o da a a e needed,
which means longe pe iods o ime need o be measu ed. Secondly, da a owne ship is an issue.
E en hough machines log da a, he e is no access o he da a, da a a e oo cos ly o is owned by
ano he company [
10
]. Thi dly, da a a e no labelled [
10
] o is labelled inconsis en ly. Ope a o s use
Appl. Sci. 2020,10, 8348 3 o 19
di e en nomencla u e o epo all mal unc ions unde he same gene ic e o code. Tha makes i
di icul o lea n wi hin a single company, le alone o lea n om ( he da a o ) ‘pee ’ companies.
Wha is needed is a s uc u ed app oach o ga he da a in a goal-o ien ed way. The c oss-indus y
s anda d p ocess o da a mining (CRISP-DM) [
11
] is a good s a ing poin , bu i desc ibes common
app oaches. The e o e, specialisa ion owa ds speci ic p edic i e main enance p oblems o domains
may be aluable. Gi en an objec i e, knowing wha o measu e is a challenge, knowing how o
measu e is e en mo e challenging.
To o e come his challenge decision suppo ools which aid in he selec ion o an op imal senso
s a egy can be used. Fo ins ance, op imal senso placemen can be ound by gene ic algo i hms [
12
]
o by using ini e elemen models [
13
]. Senso cos s is ano he ac o which can be op imised [
14
].
O he elemen s an op imal sensing s a egy decision suppo ool should ake in o accoun a e cos s
o asse /componen eplacemen , expec ed aul s and equi ed accu acy o he condi ion moni o ing
sys em. Ano he ac o he decision suppo ool should ake in o accoun is he goal o moni o ing,
o ins ance wea o a igue. To ou knowledge no esea ch has been done in o designing an o e a ching
and uni ying op imal sensing s a egy decision suppo ool.
Mo eo e , no el measu ing echnologies may appea which we a e no e en awa e o . Fo ins ance,
senso s on ain axle boxes o enable he moni o ing o insula ed ail junc ions [
15
]. O mobile phones
o commu e s could be used, so-called pa icipa o y sensing [
16
]. Fo ins ance, mobile phones ha e
been success ully used o moni o oad condi ions [
17
]. So new sensing echnologies may show up o
p edic i e main enance as well.
We see h ee main oppo uni ies, i.e., challenges o da a acquisi ion in p edic i e main enance.
Fi s , he use o senso usion, in pa icula he combina ion o emo e sensing (lida , sa elli e,
ada , sound-a ays, e c.) wi h in-si u senso s [
18
]. Senso usion may compensa e o da a gaps
o a single senso , leading o new sensing app oaches by combining senso s a di e en dis ances.
Second, he au oma ic con ex de ec ion o senso da a. Fo ins ance, a b idge is supposed o expand
du ing ho wea he , bu i he same expansion happens du ing cold wea he his could indica e
an anomaly. The challenge is o au oma ically de ec he con ex (si ua ional awa eness). Thi d,
me hods o e ec i e c oss-company da a in e ope abili y a e lacking [
19
]. These me hods make i
possible o enla ge da a se s o ge he size needed o machine lea ning. In pa icula his includes he
de ini ion o da a quali y: which quali y o da a is needed o which decision and how do we de ine
his quali y?
2.2. Da a P ocessing and Diagnos ics
Au oma ed da a alida ion and co ec ion o p edic i e main enance equi es me hods ha
wo k unde ealis ic assump ions. Wi hin s a is ics and machine lea ning, many di e en echniques
ha e been de eloped o dealing wi h missing da a [
20
–
22
]. Mos exis ing echniques ely on he
missing comple ely a andom (MCAR) assump ion, which does no apply o he ypical senso da a
ele an o p edic i e main enance. Recen app oaches based on Gaussian copulas [
23
,
24
] can a leas
handle he missing a andom (MAR) assump ion, in which whe he o no a da a poin is missing
may depend on he alues o o he a iables. A key challenge is o de elop echniques ha can u he
elax hese assump ions and e icien ly handle s eaming big da a, while a he same ime iden i ying
and co ec ing o ou lie s. Missing alue impu a ion me hods based on low- ank ma ix comple ion
such as [
25
,
26
] p o ide a good s a ing poin : hey a e compu a ionally e icien and hei implici
p ojec ion o high-dimensional da a in o a lowe -dimensional space na u ally acili a es he obus
de ec ion o ou lie s [27,28].
Moni o ing is an essen ial pa o condi ion based main enance, since moni o ing he condi ion
o sys ems allows he ea ly iden i ica ion o imminen ailu es. Cu en moni o ing me hods
a e no ye sui able o au oma ed use, since hey ail when he e is no labelled aining da a,
canno handle high-dimensional da a s eams, do no adap o da a a i ing a di e en ime scales
o do no ake in o accoun in e nal dependencies [
29
] and a e no capable o making use o physical
Appl. Sci. 2020,10, 8348 4 o 19
models. Reg ession-based moni o ing me hods ha e ecen ly been ex ended o ob ain adap i e
de ec ion h esholds in high-dimensional se ings [
30
]. A i s a emp o de elop sel -s a ing
eg ession-based moni o ing me hods ha do no equi ed labelled aining da a has been p esen ed
in [
31
]. Pu ely s a is ical app oaches ha e he ad an age o p o iding pe o mance gua an ees,
bu hey a e di icul o au oma e. A p omising ecen app oach o o e come his, is o use deep
lea ning o co ec o in e nal dependencies and use s a is ical app oaches o moni o ing [
32
].
The P imaVe a p ojec will build upon hese app oaches by de eloping au oma ed app oaches wi h
gua an eed pe o mance ha wo k in ealis ic indus ial se ings. In addi ion o hese da a-d i en
condi ion moni o ing echniques, also mo e physics-based s uc u al heal h moni o ing echniques
will be de eloped. These echniques ypically u ilise he dynamic esponse o sys ems and s uc u es
(e.g., ib a ions) o de ec and assess he p esence, loca ion and se e i y o damage [33].
To success ully design main enance in e en ions, i is essen ial o unde s and why sys ems
ail. The apidly g owing ield o causal in e ence (see, e.g., he ecen bes selle [
34
]) he e may
p o ide a solu ion. So-called ans e en opy [
35
] can be used o es ima e he di ec ed ans e
o in o ma ion be ween he ime se ies o wo a iables, e.g., om senso s a di e en pa s o a
li hog aphic machine [
36
]. Causal disco e y me hods [
37
,
38
] aim o un a el he causal s uc u e
unde lying he in e ac ions be ween many di e en a iables om pu ely obse a ional da a. F om a
me hodological poin o iew, a key challenge is o in eg a e hese wo app oaches o go om pai wise
measu es o causal in o ma ion low o a g aphical s uc u e ha can be e icien ly que ied o ind he
oo causes o speci ic ailu es.
Whe eas causal in e ence has been success ully applied in a ious scien i ic domains
(
e.g., clima e esea ch [39]
, neu oscience [
40
], p o eomics [
41
], psychology [
42
]), i s applica ion in
indus ial se ings is la gely unp eceden ed. A me hodological challenge he e is o es ima e ans e
en opy in indus ial se ings.
2.3. P ognos ics
The aim o p ognos ics is o de elop accu a e algo i hms o p edic he u u e ailu es o
componen s and sys ems. The p ognos ics s ep ollows he da a p ocessing s ep and quan i ies
ele an key pe o mance indica o s (KPI), such as he emaining use ul li e (RUL), ime o i s ailu e,
a ailabili y and eliabili y. Al hough a lo o esea ch has al eady been done in his ield, s ill se e al
majo challenges emain. The i s challenge is he gap be ween componen and sys em le el. Mos o
he me hods a ailable in li e a u e p edic ailu es on a componen le el, e.g., o bea ings [
43
], ail [
44
]
o ehicle acks [
45
]. Howe e , asse owne s a e in e es ed in he a ailabili y and expec ed ailu e o
he comple e sys em [
46
]. As de eloping sepa a e models o all componen s in a sys em s ill akes oo
much ime and e o , solu ions ha e o be ound in ei he p edic ing sys em le el ailu es om only a
limi ed numbe o (c i ical) componen models, o in speeding up he componen model de elopmen
p ocess. In he o me case, he selec ion o hese c i ical componen s, especially o la ge and complex
sys ems, is no i ial and equi es a en ion. The second challenge is ha many p edic i e models
hea ily depend on a la ge and comple e se o ailu e da a. As o well-main ained c i ical sys ems
ailu es a e by de ini ion a e, such da a se s a e o en no a ailable. This means ha da a-d i en models
mus be combined wi h domain knowledge o physics-based p ognos ic me hods [
47
]. This ela es
o he hi d challenge: only a small numbe o expe s possess de ailed knowledge on he ailu e
beha iou o componen s, which is also e y applica ion-speci ic. This makes i di icul o inco po a e
ha knowledge in gene ic p ognos ics ools. Au oma ion o he ailu e o oo cause analysis would
make his knowledge mo e accessible. The ou h challenge in p ognos ics is ha ac ual applica ion o
he me hods p oposed in scien i ic li e a u e in indus ial p ac ice appea s o be a he limi ed [
48
].
The main eason is ha companies s uggle o de e mine which app oach i s wi h hei ambi ion
and hei da a and knowledge ma u i y. The inal challenge is human ac o ela ed: enginee s a e
ypically eluc an o adop ad ice o p edic ions om ‘black box’ p ognos ic ools. Especially ully
da a-d i en and AI-based me hods a e ha d o comp ehend. Adding explainabili y [
49
] o hese kind
Appl. Sci. 2020,10, 8348 5 o 19
o me hods migh assis in inc easing us in he p edic ions. To summa ise, p ognos ic me hods a e
s ill conside ed o ha e high po en ial in p edic i e main enance, bu wide applica ion in indus y is
s ill hinde ed by bo h echnical and o ganisa ional challenges.
2.4. Main enance and Logis ics Op imisa ion
The easies way o plan main enance is o pe o m i upon ailu e, i.e., pe o m co ec i e
main enance. Howe e , his leads o many ailu es and high down ime. Fo decades now,
mos o ganisa ions ha e used some o m o p e en i e main enance: pe iodic main enance.
Main enance is hen igge ed by, o example, unning ime, calenda ime o numbe o ake-o s o an
ae oplane. The i s models we e p oposed o e 60 yea s ago by Ba low and Hun e [
50
]. Nowadays,
p edic i e main enance is an eme ging end.
Fo p edic i e main enance, in o ma ion is used ha esul s om da a acquisi ion, da a p ocessing
and diagnos ics, and p ognos ics, such as RUL es ima es o ailu e p obabili ies. Typically, as asse ge s
olde , he RUL es ima e goes down and he ailu e p obabili y goes up. I hese es ima es would be
pe ec , main enance could be pe o med exac ly be o e b eakdown. Howe e , es ima es a e impe ec
and an economic ade-o needs o be made. Pe o ming p e en i e main enance oo ea ly leads o
unnecessa y down- ime. Pe o ming p e en i e main enance oo la e leads o co ec i e main enance,
which is ypically much mo e cos ly since he main enance has o be pe o med unde high ime
p essu e, leading o high logis ics cos s o ge a se ice enginee wi h he igh pa s and ools a
he asse . Fu he mo e, down ime o a c i ical componen causes he comple e asse o be down,
which implies high down ime cos s o i s owne . This means ha he e is an economically op imal
momen o pe o m main enance ha inco po a es hese cos s and he p obabili y o ailu e o RUL
es ima e. This op imisa ion is u he complica ed because asse s con ain many (c i ical) componen s,
and g ouping main enance leads o ewe dis up ions o he cus ome and lowe logis ics cos s.
Because i is o key impo ance o pe o m main enance a he igh ime, he e has been a lo o
esea ch on making op imal p edic i e main enance decisions( o ecen e iews,
see, e.g., [5,51]).
Howe e , mos o he esea ch has been on single-i em p oblems (i.e., one ype o componen ).
Excep ions, so pape s ocusing on mul i-i em p oblems, a e hose o Zhu
[52]
and A s and Bas en
[53]
.
The e has been some esea ch on in eg a ing main enance and he se ice logis ics needed o ha e
he igh pa s, people, and ools a ailable a he momen main enance is planned [
54
,
55
] and on he
usage o condi ion moni o ing in o ma ion o adap ope a ions [
56
]. Fu he in eg a ion o he opics
o ope a ions, main enance, and se ice logis ics is equi ed. Ano he ending esea ch a ea is making
decisions wi h limi ed in o ma ion. Since he p ognos ics and o he in o ma ion a e o en a om
pe ec , models and decision making need o ake hese impe ec ions in o accoun . One way o do
ha is by modelling p oblems wi h pa ially obse able Ma ko decision p ocesses [
57
]. Such models
a e o en ha d o sol e, bu esea ch on sol ing such models is ongoing [58].
2.5. Asse Managemen and O ganisa ional Fac o s
P edic i e main enance is an ac i e esea ch a ea ha has seen signi ican p og ess o e he pas
decade, bo h in indus y and in academia. P og ess is much ela ed o ad ancemen s in he a ea o
big da a analy ics [
59
]. While many co e building blocks o p edic i e main enance (such as senso
echnology, ailu e p edic ion me hods, and op imisa ion echniques) exis , cu en solu ions ocus
on indi idual s eps in he p edic i e main enance cycle and only wo k o e y speci ic se ings as
discussed in he in oduc o y chap e o his pape . De eloping ad anced main enance echniques is
he e o e only use ul i hey a e well in eg a ed in o an o ganisa ion [60].
A quo e o a Main enance Enginee a he Ne he lands Railways who ecen ly s udied he use o
p edic i e s a egies illus a es hese o ganisa ional challenges:
“P e en i e wa e illing based on eal- ime wa e le el da a and a p edic i e model seems o be an
app op ia e main enance s a egy; howe e , his equi es he dynamic usage o human esou ces and

Appl. Sci. 2020,10, 8348 6 o 19
illing s a ions . . . T ains mo e, making he logis ic puzzle mo e complica ed . . . Ou o e all goal is o
maximise he a ailabili y o ains wi h unc ioning oile s in a cos -e ec i e way.”
T adi ionally o ganisa ional aspec s ega ding he implemen a ion o da a-d i en main enance
ha e been men ioned by o he au ho s and ha e o en been neglec ed [
48
,
60
–
63
]. The e o e,
he P imaVe a p ojec speci ically s udies he impac o da a-d i en main enance on o ganisa ional
p ocesses whe e da a-d i en main enance is being in oduced. P ocedu es o he e ec i e
implemen a ion o da a-d i en main enance sys ems wi hin o ganisa ions need o be designed in a
imely way o allow e ec i e use o i s p edic ions in ope a ional main enance planning p ocesses.
Fu he mo e, ea lie esea ch shows ha he implemen a ion o p edic i e main enance should
include ambi ion le els, a ailable da a [
64
] and a i o p edic i e main enance wi h he o ganisa ional
ma u i y o he o ganisa ion [
60
]. The ollowing o ganisa ional in e aces ha e been iden i ied by [
60
]:
s a egy and goals, decisions, s uc u e, budge and capaci y, and documen a ion. I can be deba ed
ha ea ly in eg a ed decision making is needed o e alua e he impac on hese in e aces.
Because asse managemen is a mul i-disciplina y discipline, he o ganisa ional impac s expec ed
by he in oduc ion o da a-d i en main enance sys ems on he a o emen ioned in e aces should
he e o e be app oached om mul iple pe spec i es. The pe spec i es men ioned by [
65
], e.g., echnical,
economic, comme cial, compliance, and o ganisa ional aspec s seem ele an o be used he e,
especially because asse managemen aspec s a e a ely limi ed o a one-dimensional pe spec i e.
The mos c i ical o ganisa ional impac s should he e o e ideally be iden i ied and assessed be o e
he in oduc ion o da a-d i en main enance by s udying he use o he a o emen ioned pe spec i es
in his speci ic asse managemen a ea. Because o he complexi y o he associa ed sys ems, p ocesses,
and people he e will always emain a numbe o o ganisa ional decisions ha need o be iden i ied
and add essed be o e da a-d i en main enance o indi idual componen s can ac ually be implemen ed.
As [
66
] poin ed ou he e a e always ade-o s be ween main enance cos s, a ailabili y and e iciency
in (mul i-componen ) sys ems.
I can be a gued based on he ou comes o he wo k o Koochaki [
66
] ha o ganisa ional p ocesses
need o become mo e lexible o make da a-d i en main enance on a mul i-componen sys em mo e
easible. The e o e, he P imaVe a p ojec will also in es iga e how o ganisa ional eadiness and
esilience in p ocesses can be de eloped be o e o du ing he in oduc ion o hese sys ems. The use o
high- eliabili y heo y and an i- agili y in o ganisa ions can be seen as eme ging ields [
67
] besides
he needed a en ion o cul u al aspec s [60].
Fo he de elopmen o app op ia e decision making suppo ools an i e a i e design science
esea ch (DSR) app oach [
68
] is en isioned in which a e ac s a e i e a i ely e alua ed and imp o ed.
A DSR s a egy ocuses on de eloping a e ac s as well as knowledge c ea ion, and aims o p oduce
imp o emen s based on a ho ough unde s anding o p oblems o oppo uni ies [
68
]. The e o e,
he ou come o DSR is no only ele an o he p ac ical applica ion domain, bu is also explici ly aimed
a he c ea ion o heo e ical knowledge [69].
2.6. Human Fac o s
Human beings a e c i ical o he unc ioning and pe o mance o he majo i y o ope a ing sys ems.
Howe e , human beha iou adi ionally has been igno ed in he ield o ope a ion managemen
(OM). Tha is, mos models in OM assume ha agen s who pa icipa e in ope a ing p ocesses a e
ei he ully a ional o can be induced o beha e a ionally [
70
,
71
]. Mo e speci ically, hese models
assume ha people ha e s able p e e ences, a e no a ec ed by cogni i e biases o emo ions, and ha e
he abili y o dis ega d i ele an in o ma ion by only esponding o ele an in o ma ion when
making decisions [
72
]. The eme ging ield o Beha iou al Ope a ions Managemen depa s om
hese ( a he un ealis ic) assump ions by acknowledging ha human decision-make s a e guided
by emo ions, cogni i e biases o i ele an si ua ional cues ha may a ec he adop ion and usage
o ope a ing sys ems [
73
]. Mo e speci ically, ans o ming main enance sys ems and ope a ions
in o ones ha ely on da a-d i en echnologies b ing many challenges. One impo an challenge
Appl. Sci. 2020,10, 8348 7 o 19
conce ns he design o (da a-d i en) main enance sys ems ha o ganisa ional membe s a e willing
o us and use [
74
,
75
]. Tha is, in o de o success ully in eg a e hese p omising echnologies in o
o ganisa ions, i is o c i ical impo ance o unde s and when and why use s a e hesi an o adop
hese new echnologies in hei daily wo king ou ine and how we can s imula e i s e ec i e usage.
As such, he goal and no el y o P imaVe a is o de elop key insigh s in o (i) wha ac o s impac s a
pe son’s accep ance and use o da a-d i en ailu e p edic ions and main enance ecommenda ions and
(ii) how o e ec i ely combine human judgemen wi h he solu ion o a sys em. These insigh s will
be used o design no el, use -cen ed main enance ools ha make he use –sys em in e ac ion mo e
e ec i e and e icien .
3. Obs acles o O e come
To eap he ui s o p edic i e main enance and le e age indi idual building blocks in o an
e ec i e solu ion, he P imaVe a p ojec has iden i ied ou c oss-cu ing obs acles ha need o
be o e come. These obs acles ha e been es ablished based on ou own expe in e iews om
people wi hin academia and indus y and a e backed by ecen insigh s om majo consul ancy
i ms [76,77].
The obs acles ha hus a ha e hinde ed e ec i e solu ions a e lack o o ches a ion,
lack o au oma ion, da a unce ain y and he ole o human and o ganisa ional ac o s. The jus i ica ion
o hese obs acles is suppo ed by an empi ical Delphi-based scena io planning s udy conduc ed
wi hin he a ea o main enance in digi alised manu ac u ing [
78
]. Each o hese obs acles a e de ailed
in he ollowing sec ions.
3.1. O ches a ion
Cu en p edic i e main enance solu ions o en ocus on a single s ep in he p edic i e
main enance chain, wi h poo alignmen o he es o he wo k low. This is subop imal, since locally
op imal solu ions do no usually lead o o e all op imal solu ions. Thus, e ec i e main enance
equi es no el op imisa ion echniques ha wo k ac oss di e en agg ega ion le els. In pa icula ,
asse managemen in ol es supply chain op imisa ion, o ches a ing he planning o main enance
pe sonnel, equipmen and g oups o asse s.
Bok an z e al. en ision ha e ec i e main enance will lead o op imised pe o mance o en i e
manu ac u ing sys ems [
78
]. To achie e his he challenge is o de elop me hods and algo i hms
which a e use ul in p ac ice [
78
]. Implemen ing p edic i e main enance solu ions which ocus on a
single ailu e mode a e non-op imal, solu ions mus be implemen ed which conside he in e ac ion
o componen s wi h hei ope a ing en i onmen [
79
]. Kippe e al. sugges ha esea che s need o
de elop s udies o imp o e he unde s anding o how Indus y 4.0 echnologies and concep s impac
p ocesses, p oduc s and se ices [80].
3.2. Au oma ion
Cu en applica ions o p edic i e main enance usually consis o a la ge numbe o non-au oma ed
p ocedu es. This is no only ine icien , bu also e o p one. Au oma ing hese s eps in o sys ema ic
p ocedu es is challenging, because hey in ol e a weal h o domain knowledge. In pa icula , accu a e,
scalable and obus algo i hms o da a cleaning, causal disco e y o ailu es and oo cause analysis
a e cu en ly lacking, and he same holds o p edic ion algo i hms o so wa e and elec onics, as well
as o algo i hms o op imise he supply chain logis ics.
Bok an z e al. unde line his obs acle by no ing ha de eloping main enance managemen
sys ems ha au oma ically ans o m big da a in o decision suppo is s ill challenging [
78
]. Kippe e al.
ecommends u u e esea ch should be ca ied ou in o de o de elop amewo ks o deploying
Indus y 4.0 in eal applica ions, such as p edic i e main enance, no only in la ge companies bu also
in Small Medium En e p ises (SMEs) [80].
Appl. Sci. 2020,10, 8348 8 o 19
3.3. Da a Unce ain y
Da a om senso s o o he sou ces is o en inaccu a e o incomple e. Ob aining accu a e
p ognos ics and main enance decisions despi e impe ec and unce ain da a equi es sophis ica ed
me hods ha a e capable o handling eal wo ld unce ain ies [
81
]. Since unce ain ies p opaga e
along he p edic i e main enance wo k low, hese echniques play a ole in each s ep o he p edic i e
main enance cycle. Me hods which e alua e he e ec i eness and accu acy o p edic i e main enance
solu ions wi h ega ds o unce ain y a e equi ed [81].
3.4. Human and O ganisa ional Fac o s
The ansi ion owa ds he Indus y 4.0 equi es o ganisa ions o embed he da a-d i en
cul u e in o hei wo k low. A key issue is he us in da a-d i en decision suppo ools:
main enance decisions ha a e au oma ically compu ed by ools mus be ac ed upon by main enance
enginee s. This equi es a use -cen ic design o hese decision suppo ools. In addi ion, he p ojec
eam esponsible o success ully deploying a p edic i e main enance solu ion is o en con on ed wi h
eluc ance and ese a ions [
10
]. The lack o communica ion be ween be ween heo y de elope s and
p ac i ione s in he a ea o eliabili y and main enance is also an issue [82].
Besides he igh p esen a ion o in o ma ion, he igh p ocess in o ma ion needs o be made
a ailable a he igh momen o allow da a-d i en main enance ac i i ies. O en he e is limi ed
in o ma ion a ailable on expec ed o ganisa ional impac s o da a-d i en main enance ac ions.
4. P edic i e Main enance P ocess Model
To o e come he be o e men ioned obs acles and o o ches a e he di e en s eps in he p edic i e
main enance wo k low a gene ic applicable p ocess model is p oposed o acili a e his (Figu e 1).
The p oposed model is simila o exis ing models [
83
–
87
], which a e in essence all based on he gene ic
model o Ja dine e al. [
82
]. The model o Ja dine e al. has h ee dis inc s ages; da a acquisi ion,
da a p ocessing and main enance decision-making. As he goal is o de ine a gene ically applicable
p ocess model o p edic i e main enance, no only should i be applicable a sys em le el bu also
a lee le el [
88
,
89
]. To ensu e his gene ici y, he p oposed model will also be based on he gene ic
model o Ja dine e al. To highligh he signi icance o diagnos ics and p ognos ics wi hin p edic i e
main enance, he p oposed model explici ly b eaks down he da a p ocessing s ep in o hese wo
elemen s. The impo ance o human and o ganisa ion ac o s is commonly o e looked by enginee ing
disciplines [
48
,
61
–
63
] bu is well oo ed wi hin in o ma ion sys ems esea ch [
90
,
91
]. Technology and
beha iou a e no dicho omous [
90
], he e o e human and o ganisa ional ac o s ha e been added
o he p oposed model. This elemen is placed a he e y cen e o he model as i a ec s all o he
s ages o he model. I is exac ly his addi ion which se s he p oposed model apa om he p e iously
p oposed models. The usabili y and applicabili y o he p oposed model will be e alua ed by applying
i o he demons a o s desc ibed in Sec ion 5.1.
The gene ic model consis s o i e s ages: (1) da a a e acqui ed om asse s using senso s o
o he sou ces, (2) hese da a a e hen p ocessed and u ned in o meaning ul diagnos ic in o ma ion
h ough da a selec ion, cleaning and in e p e a ion, (3) om his in o ma ion p edic ions a e made
abou he sys em’s heal h (p ognos ics), (4) based on hese p ognos ics, main enance and associa ed
logis ics a e op imised, (5) all in o ma ion has o be inco po a ed in o a s a egic asse managemen
plan. Decisions lis ed in he asse managemen plan a e ans o med in o ac ions which will a ec he
asse being managed, hence closing he cycle. An asse managemen plan documen s he ac i i ies,
esou ces and imescales equi ed o achie e he o ganisa ion’s asse managemen objec i es o an
indi idual asse o g oup o asse s [
92
]. No e ha each o he s ages ela e o one o mo e o he be o e
men ioned obs acles o p edic i e main enance o o e come.
The s a ing poin wi hin he p edic i e main enance cycle depends on he mo i a ion o asse
managemen [
93
]. This mo i a ion can be ini ia ed by a echnology push; exis ing echnology is
Appl. Sci. 2020,10, 8348 9 o 19
a ailable which needs o be managemen , in his case he cycle would s a wi h da a acquisi ion.
On he o he hand mo i a ion can be ini ia ed by a decision pull; he e is a ce ain economic necessi y,
in his case he cycle would s a wi h an asse managemen plan.
Human and o ganisa ional
ac o s
Asse s
Da a
In o ma ion
Pe o mance
indica o s
Asse managemen
plan
Da a acquisi ion
Da a p ocessing
and diagnosis
P ognos ics
Main enance and
logis ics op imisa ion
Decisions
Figu e 1. P edic i e main enance managemen p ocess model.
4.1. P imaVe a App oach
The subsequen sec ions will ocus on each indi idual s age o he p oposed p ocess model and
will desc ibe he en isaged scien i ic ou come o his s age. Once again i should be s essed ha he
p ojec ’s main endea ou is o en ol a holis ic, c oss-sec o al app oach, hus explici ly add essing he
obse ed obs acle o a lack o o ches a ion.
4.1.1. Da a Acquisi ion
In p ac ical se ings selec ion o sui able senso s o implemen ing a p edic i e main enance
solu ion pose a challenge [
10
]. To o e come his, a decision suppo ool will be ealised ha ad ises
on he mos app op ia e sensing echniques, spa ial senso placemen and op imal sensing s a egy o
moni o an asse . Though he e is a lo o wo k on op imal senso placemen [
12
,
13
,
94
], an o e a ching
decision suppo ool which akes all ace s o p edic i e main enance in o accoun is s ill lacking.
Inpu o such a decision suppo ool will include c i ical componen s oge he wi h hei ailu e
modes, equi ed accu acy and esolu ion, cos ac o s and expe domain knowledge. Implemen ing
an op imal sensing s a egy will aid in he mi iga ion o da a unce ain y.
One o he case s udies being analysed in he P imaVe a p ojec is a sludge d edge ’s p opulsion
sys em. A limi ed amoun o eco ded ailu e da a is a ailable o his sys em, only eigh clea ly labelled
eminen ailu es a e p esen . In o de o ob ain accu a e p ognos ic models, mo e ailu e da a a e
equi ed. The e o e, a ious me hods o acqui e mo e ailu e da a will be e alua ed. Fi s me hod is
a model based app oach o ga he mo e ailu e da a. A compu a ional whi e box model is c ea ed
based on a quali a i e unc ional decomposi ion o he sys em. Second me hod is he use o a scaled
physical model o a p opulsion sys ems which pu pose ully has damaged componen s ins alled such
as aul y bea ings. Thi d me hod is he use o public a ailable da a se s om simila sys ems such as
he Machine y Faul Da abase [
95
] o e alua e he easibili y o ans e lea ning. T ans e lea ning
allows he domains, asks and dis ibu ions o be di e en o aining and es ing [96].
Appl. Sci. 2020,10, 8348 16 o 19
48.
Tiddens, W.W.; B aaksma, A.J.J.; Tinga, T. Towa ds In o med Main enance Decision Making: Guiding he
Applica ion o Ad anced Main enance Analyses. In Op imum Decision Making in Asse Managemen ;
IGI Global: He shey, PA, USA, 2016; Chap e 13, pp. 288–309. [C ossRe ]
49.
en Zeldam, S.; de Jong, A.; Loende sloo , R.; Tinga, T. Au oma ed Failu e Diagnosis in A ia ion
Main enance Using eXplainable A i icial In elligence (XAI). In P oceedings o he Eu opean Con e ence
o he PHM Socie y, 4 h Eu opean Con e ence o he P ognos ics and Heal h Managemen Socie y,
PHME 2018—Mun gebouw U ech , U ech , The Ne he lands, 3–6 July 2018; Volume 4.
50. Ba low, R.; Hun e , L. Op imum p e en i e main enance policies. Ope . Res. 1960,8, 90–100. [C ossRe ]
51.
Olde Keize , M.C.; Flappe , S.D.P.; Teun e , R.H. Condi ion-based main enance policies o sys ems wi h
mul iple dependen componen s: A e iew. Eu . J. Ope . Res. 2017,261, 405–420. [C ossRe ]
52.
Zhu, Q. Main enance Op imiza ion o Mul i-Componen Sys ems unde Condi ion Moni o ing.
Ph.D. Thesis, Eindho en Uni e si y o Technology, Eindho en, The Ne he lands, 2015.
53.
A s, J.; Bas en, R. Design o mul i-componen pe iodic main enance p og ams wi h single-componen
models. IISE T ans. 2018,50, 606–615. [C ossRe ]
54.
Van Ho enbeek, A.; Bu é, J.; Ca ysse, D.; Pin elon, L.; Vans eenwegen, P. Join main enance and in en o y
op imiza ion sys ems: A e iew. In . J. P od. Econ. 2013,143, 499–508. [C ossRe ]
55.
Bas en, R.; Ryan, J. The alue o main enance delay lexibili y o imp o ed spa e pa s in en o y
managemen . Eu . J. Ope . Res. 2019,278, 646–657. [C ossRe ]
56.
ui he B oek, M.A.; Teun e , R.H.; de Jonge, B.; Veldman, J.; Van Fo ees , N.D. Condi ion-based p oduc ion
planning: Adjus ing p oduc ion a es o balance ou pu and ailu e isk. Manu . Se . Ope . Manag.
2019
,22,
645–867. [C ossRe ]
57.
Ka aba˘g, O.; E uguz, A.S.; Bas en, R. In eg a ed op imiza ion o main enance in e en ions and spa e pa
selec ion o a pa ially obse able mul i-componen sys em. Reliab. Eng. Sys . Sa .
2020
,200, 106955.
[C ossRe ]
58.
Ca , S.; Jansen, N.; Wimme , R.; Se ban, A.C.; Becke , B.; Topcu, U. Coun e example-Guided S a egy
Imp o emen o POMDPs Using Recu en Neu al Ne wo ks. a Xi 2019, a Xi :1903.08428.
59.
Ra deep, K.; Adi hya, T.; Sa bjee , S.; Ma ine i, A. Big Da a Analy ics o Main aining T anspo a ion
Sys ems. In T anspo a ion Sys ems; Sp inge : Be lin/Heidelbe g, Ge many, 2004; Chap e 6, pp. 85-1–85-11.
[C ossRe ]
60.
an de Ke kho , R. I ’s abou Time: Managing Implemen a ion Dynamics o Condi ion-Based Main enance.
Ph.D. Thesis, Uni e si y o Tilbu g, Tilbu g, The Ne he lands, 2020.
61.
Ga g, A.; Deshmukh, S.G. Main enance managemen : Li e a u e e iew and di ec ions. J. Qual. Main . Eng.
2006,12, 205–238. [C ossRe ]
62.
an de Ke kho , R.; Akke mans, H.A. Noo de ha en, N. Knowledge Los in Da a: O ganiza ional
Impedimen s o Condi ion-Based Main enance in he P ocess Indus y. In Logis ics and Supply Chain Inno a ion;
Sp inge : Be line/Heidelbe g, Ge many, 2015; p. 19.
63.
Veldman, J.; Wo mann, H.; Klingenbe g, W. Typology o condi ion based main enance. J. Qual. Main . Eng.
2011,17, 183–202. [C ossRe ]
64.
Tiddens, W.W.; B aaksma, A.J.J.; Tinga, T. Selec ing Sui able Candida es o P edic i e Main enance. In . J.
P ogn. Heal h Manag. 2018,9, 20.
65.
Rui enbu g, R.J.; B aaksma, A.J.J. Mi iga ing change in he goals and con ex o capi al asse s: Design o he
Li e ime Impac Iden i ica ion Analysis. CIRP J. Manu . Sci. Technol. 2017,17, 50–59. [C ossRe ]
66.
Koochaki, J. CBM in Mul i-Componen Sys ems. Ph.D. Thesis, Uni e si y o G oningen, G oningen,
The Ne he lands, 2012. A ailable online: h ps://www. ug.nl/ (accessed on 31 July 2020).
67.
Ma ine i, A.; Moe man, J.J.; an Dongen, L.A. S o y elling as a s a egy in managing complex sys ems:
Using an i agili y o handling an unce ain u u e in eliabili y. Sa . Reliab. 2017,37, 233–247. [C ossRe ]
68.
an Aken, J.E.; Chand aseka an, A.; Halman, J. Conduc ing and publishing design science esea ch:
Inaugu al essay o he design science depa men o he Jou nal o Ope a ions Managemen . J. Ope . Manag.
2016,47–48, 1–8. [C ossRe ]
69.
Wie inga, R.J. Design science me hodology: Fo in o ma ion sys ems and so wa e enginee ing. In Design
Science Me hodology: Fo In o ma ion Sys ems and So wa e Enginee ing; Sp inge : Be lin/Heidelbe g, Ge many,
2014; pp. 1–332. [C ossRe ]

Appl. Sci. 2020,10, 8348 17 o 19
70.
Bendoly, E.; Donohue, K.; Schul z, K.L. Beha io in ope a ions managemen : Assessing ecen indings and
e isi ing old assump ions. J. Ope . Manag. 2006,24, 737–752. [C ossRe ]
71.
Gino, F.; Pisano, G. Towa d a Theo y o Beha io al Ope a ions. Manu . Se . Ope . Manag.
2008
,10, 676–691.
[C ossRe ]
72.
on Neumann, J.; Mo gens e n, O. Theo y o Games and Economic Beha io . J. Philos.
1945
,42, 550.
[C ossRe ]
73.
Donohue, K.; Öze , Ö.; Zheng, Y. Beha io al Ope a ions: Pas , P esen , and Fu u e. Manu . Se . Ope . Manag.
2020,22, 191–202. [C ossRe ]
74.
Die o s , B.J.; Simmons, J.P.; Massey, C. Algo i hm a e sion: People e oneously a oid algo i hms a e
seeing hem e . J. Exp. Psychol. Gen. 2015,144, 114–126. [C ossRe ]
75.
E endi´c, E.; Van de Calseyde, P.P.; E ans, A.M. Slow esponse imes unde mine us in algo i hmic (bu no
human) p edic ions. O gan. Beha . Hum. Decis. P ocess. 2020,157, 103–114. [C ossRe ]
76.
Manyika, J.; Chui, M.; Bisson, P.; Woe zel, J.; Dobbs, R.; Bughin, J.; Aha on, D. The In e ne o Things: Mapping
he Value beyond he Hype; Technical Repo ; McKinsey Global Ins i u e: New Yo k, NY, USA, 2015.
77.
Haa man, M.; Mulde s, M.; Vassiliadis, C. P edic i e Main enance 4.0: P edic he Unp edic able;
Technical Repo ; PwC Ne he lands and Mainno a ion: Ams e dam, The Ne he lands, 2017.
78.
Bok an z, J.; Skoogh, A.; Be lin, C.; S ah e, J. Main enance in digi alised manu ac u ing: Delphi-based
scena ios o 2030. In . J. P od. Econ. 2017,191, 154–169. [C ossRe ]
79.
Heng, A.; Zhang, S.; Tan, A.C.; Ma hew, J. Ro a ing machine y p ognos ics: S a e o he a , challenges and
oppo uni ies. Mech. Sys . Signal P ocess. 2009,23, 724–739. [C ossRe ]
80.
Kippe , L.M.; Fu s enau, L.B.; Hoppe, D.; F ozza, R.; Iepsen, S. Scopus scien i ic mapping p oduc ion in
indus y 4.0 (2011–2018): A bibliome ic analysis. In . J. P od. Res. 2020,58, 1605–1627. [C ossRe ]
81.
Sun, B.; Zeng, S.; Kang, R.; Pech , M.G. Bene i s and Challenges o Sys em P ognos ics. IEEE T ans. Reliab.
2012,61, 323–335. [C ossRe ]
82.
Ja dine, A.K.; Lin, D.; Banje ic, D. A e iew on machine y diagnos ics and p ognos ics implemen ing
condi ion-based main enance. Mech. Sys . Signal P ocess. 2006,20, 1483–1510. [C ossRe ]
83.
Ja ed, K.; Gou i eau, R.; Ze houni, N.; Nec oux, P. Enabling Heal h Moni o ing App oach Based on
Vib a ion Da a o Accu a e P ognos ics. IEEE T ans. Ind. Elec on. 2015,62, 647–656. [C ossRe ]
84.
Vach se anos, G.; Goebel, K. In oduc ion o P ognos ics. 2015. A ailable online: h ps://www.phmsocie y.
o g/si es/phmsocie y.o g/ iles/PROGNOSTICS_TUTORIAL.pd (accessed on 23 No embe 2020).
85.
Vach se anos, G.J.; Vala anis, K.P. A No el App oach o In eg a ed Vehicle Heal h Managemen ; Technical Repo ;
NATO STO: B ussels, Belgium, 2018. [C ossRe ]
86.
Benede ini, O.; Baines, T.S.; Ligh oo , H.W.; G eenough, R.M. S a e-o - he-a in in eg a ed ehicle heal h
managemen . P oc. Ins . Mech. Eng. Pa G J. Ae osp. Eng. 2009,223, 157–170. [C ossRe ]
87.
Ran, Y.; Zhou, X.; Lin, P.; Wen, Y.; Deng, R. A Su ey o P edic i e Main enance: Sys ems, Pu poses and
App oaches. a Xi 2019, a Xi :eess.SP/1912.07383.
88.
Lee, J.; Jin, C.; Liu, Z.; Da a i A dakani, H. In oduc ion o Da a-D i en Me hodologies o P ognos ics
and Heal h Managemen . In P obabilis ic P ognos ics and Heal h Managemen o Ene gy Sys ems; Sp inge :
Be lin/Heidelbe g, Ge many, 2017; pp. 9–32. [C ossRe ]
89.
Jin, X.; Siegel, D.; Weiss, B.A.; Gamel, E.; Wang, W.; Lee, J.; Ni, J. The p esen s a us and u u e g ow h o
main enance in US manu ac u ing: Resul s om a pilo su ey. Manu . Re . 2016,3, 10. [C ossRe ]
90.
He ne , A.; Ma ch, S.T.; Pa k, J.; Ram, S. Design Science in In o ma ion Sys ems Resea ch. MIS Q.
2004
,
28, 75. [C ossRe ]
91.
Pe e s, K.; Tuunanen, T.; Ro henbe ge , M.A.; Cha e jee, S. A Design Science Resea ch Me hodology o
In o ma ion Sys ems Resea ch. J. Manag. In . Sys . 2007,24, 45–77. [C ossRe ]
92.
Asse Managemen —O e iew, P inciples and Te minology; S anda d, In e na ional O ganiza ion o S anda diza ion:
Gene a, Swi ze land, 2014.
93.
Tiddens, W.W.; B aaksma, J.; Tinga, T. Explo ing p edic i e main enance applica ions in indus y. J. Qual.
Main . Eng. 2020. [C ossRe ]
94.
Gup a, V.; Sha ma, M.; Thaku , N. Op imiza ion C i e ia o Op imal Placemen o Piezoelec ic Senso s
and Ac ua o s on a Sma S uc u e: A Technical Re iew. J. In ell. Ma e . Sys . S uc .
2010
,21, 1227–1243.
[C ossRe ]
Appl. Sci. 2020,10, 8348 18 o 19
95.
Ribei o, F. Machine y Faul Da abase. 2017. A ailable online: h p://www02.sm .u j.b /~o sho e/m s/
index.h ml (accessed on 11 Sep embe 2020).
96.
Pan, S.J.; Yang, Q. A Su ey on T ans e Lea ning. IEEE T ans. Knowl. Da a Eng.
2010
,22, 1345–1359.
[C ossRe ]
97.
Sanka a aman, S.; Goebel, K. Unce ain y in p ognos ics and sys ems heal h managemen . In . J. P ogn.
Heal h Manag. 2015,6.
98.
Liu, D.; Luo, Y.; Peng, Y. Unce ain y p ocessing in p ognos ics and heal h managemen : An o e iew.
In P oceedings o he IEEE 2012 P ognos ics and Sys em Heal h Managemen Con e ence (PHM-2012 Beijing),
Beijing, China, 23–25 May 2012; pp. 1–6. [C ossRe ]
99.
Ng, K.C.; Ab amson, B. Unce ain y managemen in expe sys ems. IEEE Expe
1990
,5, 29–48. [C ossRe ]
100.
an de Ho s ad, R. Random G aphs and Complex Ne wo ks; Camb idge Uni e si y P ess: Camb idge, UK, 2016.
[C ossRe ]
101.
Imbens, G.W.; Rubin, D.B. Causal In e ence o S a is ics, Social, and Biomedical Sciences; Camb idge Uni e si y
P ess: Camb idge, UK, 2015. [C ossRe ]
102.
Kene , R.S. On gene a ing high In oQ wi h Bayesian ne wo ks. Qual. Technol. Quan . Manag.
2016
,
13, 309–332. [C ossRe ]
103.
Nielsen, T.D.; Jensen, F.V. Bayesian Ne wo ks and Decision G aphs; Sp inge : New Yo k, NY, USA, 2007.
[C ossRe ]
104.
Ding, M.; Chen, Y.; B essle , S.L. G ange Causali y: Basic Theo y and Applica ion o Neu oscience.
In Handbook o Time Se ies Analysis; John Wiley & Sons, L d.: Hoboken, NJ, USA, 2006; Chap e 17, pp. 437–460.
[C ossRe ]
105.
Duan, P.; Yang, F.; Chen, T.; Shah, S.L. Di ec Causali y De ec ion ia he T ans e En opy App oach.
IEEE T ans. Con ol Sys . Technol. 2013,21, 2052–2066. [C ossRe ]
106.
Lee, J.; Wu, F.; Zhao, W.; Gha a i, M.; Liao, L.; Siegel, D. P ognos ics and heal h managemen design o
o a y machine y sys ems—Re iews, me hodology and applica ions. Mech. Sys . Signal P ocess.
2014
,
42, 314–334. [C ossRe ]
107.
Schwabache , M.; Goebel, K. A Su ey o A i icial In elligence o P ognos ics; A i icial In elligence o
P ognos ics; The AAAI P ess: Menlo Pa k, CA, USA, 2007; pp. 108–115.
108. Tinga, T. P inciples o Loads and Failu e Mechanisms; Sp inge : London, UK, 2013. [C ossRe ]
109.
Ruij e s, E.; Guck, D.; D olenga, P.; S oelinga, M. Faul main enance ees: Reliabili y cen e ed main enance
ia s a is ical model checking. In P oceedings o he 2016 Annual Reliabili y and Main ainabili y Symposium
(RAMS), Tucson, AZ, USA, 25–28 Janua y 2016; pp. 1–6.
110.
Nau a, M.; Bucu , D.; S oelinga, M. LIFT: Lea ning Faul T ees om Obse a ional Da a. In P oceedings o
he 15 h In e na ional Con e ence on Quan i a i e E alua ion o Sys ems; Lec u e No es in Compu e Science;
McI e , A., Ho á h, A., Eds.; Sp inge : Be lin/Heidelbe g, Ge many, 2018; Volume 11024, pp. 306–322.
[C ossRe ]
111.
Lina d, A.; Bucu , D.; S oelinga, M. Faul T ees om Da a: E icien Lea ning wi h an E olu iona y
Algo i hm. In P oceedings o he 5 h In e na ional Symposium on Dependable So wa e Enginee ing: Theo ies,
Tools, and Applica ions SETTA; Lec u e No es in Compu e Science; Guan, N., Ka oen, J., Sun, J., Eds.;
Sp inge : Be lin/Heidelbe g, Ge many, 2019; Volume 11951, pp. 19–37. [C ossRe ]
112.
Yamada, S.; Osaki, S. So wa e Reliabili y G ow h Modeling: Models and Applica ions. IEEE T ans.
So w. Eng. 1985,SE-11, 1431–1437. [C ossRe ]
113.
Ja ain, A.; Meh a, Y. Me ics and models o So wa e Reliabili y: A sys ema ic e iew. In P oceedings o he
In e na ional Con e ence on Issues and Challenges in In elligen Compu ing Techniques (ICICT), Ghaziabad,
India, 7–8 Feb ua y 2014; pp. 210–214. [C ossRe ]
114.
Tian, L.; Noo e, A. E olu iona y neu al ne wo k modeling o so wa e cumula i e ailu e ime p edic ion.
Reliab. Eng. Sys . Sa . 2005,87, 45–51. [C ossRe ]
115.
Li, Z. Using Da a Mining Techniques o Imp o e So wa e Reliabili y. Ph.D. Thesis, Uni e si y o Illinois,
Champaign, IL, USA, 2006.
116.
Heijblom, R.; Pos ma, W.; Na a ajan, V.; S oelinga, M. DFT Analysis Inco po a ing Spa e Pa s in Faul T ees.
In P oceedings o he 2018 Annual Reliabili y and Main ainabili y Symposium (RAMS), Reno, NV, USA,
22–25 Janua y 2018; pp. 1–7.
Appl. Sci. 2020,10, 8348 19 o 19
117.
Bas en, R.J.I.; an Hou um, G.J. Sys em-o ien ed in en o y models o spa e pa s. Su . Ope . Res.
Manag. Sci. 2014,19, 34–55. [C ossRe ]
118.
Häkkinen, L.; Hilmola, O. Li e a e ERP implemen a ion: Long- e m de elopmen o use pe cep ions o
sys em success in an a e -sales en i onmen . J. En e p. In . Manag. 2008,21, 285–310. [C ossRe ]
119.
Venka esh, V.; Bala, H. Technology Accep ance Model 3 and a Resea ch Agenda on In e en ions. Decis. Sci.
2008,39, 273–315. [C ossRe ]
120.
Venka esh, V.; Mo is, M.G.; Da is, G.B.; Da is, F.D. Use Accep ance o In o ma ion Technology: Towa d a
Uni ied View. Manag. In . Sys . Q. 2003,27. [C ossRe ]
121.
Wie s, V. The ela ionship be ween shop loo au onomy and APS implemen a ion success: E idence om
wo cases. P od. Plan. Con ol 2009,20, 576–585. [C ossRe ]
122.
Fildes, R.; Goodwin, P.; Law ence, M.; Nikolopoulos, K. E ec i e o ecas ing and judgmen al adjus men s:
An empi ical e alua ion and s a egies o imp o emen in supply-chain planning. In . J. Fo ecas .
2009
,
25, 3–23. [C ossRe ]
123. S inge , E. Ac ion Resea ch, 4 h ed.; SAGE Publica ions, Inc.: Los Angeles, CA, USA, 2013.
124. Ebe , C.; Galla do, G.; He nan es, J.; Se ano, N. De Ops. IEEE So w. 2016,33, 94–100. [C ossRe ]
125.
Boe ige , C. An In oduc ion o Docke o Rep oducible Resea ch. SIGOPS Ope . Sys . Re .
2015
,49, 71–79.
[C ossRe ]
Publishe ’s No e:
MDPI s ays neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional
a ilia ions.
c
2020 by he au ho s. Licensee MDPI, Basel, Swi ze land. This a icle is an open access
a icle dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion
(CC BY) license (h p://c ea i ecommons.o g/licenses/by/4.0/).