T u t o r i al s o n
L i f e Cy cl e Co st i n g
an d
Rel i ab i l i t y
En gi n eer i n g
Co u r se M at er i al
Cour se Inst r uct or : Pr of essor U Di nesh Kumar
Indian Inst i t ut e of M anagement Bangal or e
1. Rel i abi l i t y Mai n t enance and Logi st i c Suppor t  I nt r oduct i on 2
Chapter 1
Reliability Maintenance and Logistic
Support  Introduction
All the business of war, and indeed all the business of life, is to endeavour
to find out what you dont know from what you do.
Duke of Wel l i ngt on
1.1. INTRODUCTION
Ever si nce t he Indust r i al Revol ut i on began some 2½ cent ur i es ago,
cust omer s have demand ed bet t er , cheaper , f ast er , mor e f or l ess, t hr ou gh
gr eat er r el i abi l i t y, mai nt ai nabi l i t y and suppor t abi l i t y (RM S). As soon as
peopl e set t hemsel ves up i n busi n ess t o pr ovi de pr o duct s f or ot her s and
not j ust f or t hemsel ves, t hei r cust o mer s have al w ays want ed t o make sur e
t hey wer e not bei ng expl oi t ed and t hat t hey wer e get t i ng val ue f or money
and pr oduct s t hat woul d be f i t f or pur pose.
Today’ s cust omer s ar e no di f f er ent . Al l t hat has changed i s t hat t he
compani es have gr ow n bi gger , t h e pr oduct s have b eco me mor e
sophi st i cat ed, compl ex and expensi ve and, t he cust omer s have become
mor e demandi ng and even l ess t r ust i ng. As i n al l f or ms of evol ut i on, t he
Red Queen Syndr ome ( Lew i s, C. 1971, M at t , R., 1993) i s f or ever pr esent – i n
busi ness, as i n al l t hi ngs, you si mpl y have t o keep r u nni ng f ast er t o st and
st i l l . No mat t er how good you make som et hi ng, i t wi l l never r emai n good
enough f or l ong.
Oper at or s w ant i nf i ni t e per f or mance, at zer o l i f ecycl e cost , w i t h
100% avai l abi l i t y f r om t he day t hey t ake t o del i ver y t o t he day t h ey di spose
of i t . It i s t he t ask of t he desi gner / manuf act ur er / suppl i er / pr oducer t o get
as near as possi bl e t o t h ese ext r emes, or , at t h e ver y l east , near er t han
1. Rel i abi l i t y Mai n t enance and Logi st i c Suppor t  I nt r oduct i on 3
t hei r comp et i t or s. In many cases, how ever , i t i s not si mpl y suf f i ci ent t o t ell
t he (pot ent i al ) cust omer how w el l t hey have met t hese r eq ui r ement s,
r at her , t hey wi l l be r equi r ed t o pr oduce demonst r abl e evi dence t o
subst ant i at e t h ese cl ai ms. In t he f ol l owi ng pages, we hope t o pr ovi de you
wi t h t he t echni ques and met hodol ogi es t hat wi l l enabl e you t o do t hi s and,
t hr ough pr act i cal exampl es, expl ai n how t hey can be used.
The success of any busi ness d ep ends on t he ef f ect i veness of t he
pr ocess and t he pr oduct t hat busi ness pr oduces. Ever y pr oduct i n t hi s w or l d
i s made t o per f or m a f unct i on and ever y cust omer / user woul d l i ke her
pr oduct t o mai nt ai n i t s f unct i onal i t y unt i l has f ul f i l l ed i t s pur pose or , f ai l i ng
t hat , f or as l ong as possi bl e. If t hi s can be don e wi t h t he mi ni mum of
mai nt enance but , when t her e i s a need f or mai nt enance, t hat t hi s can be
done i n t he mi ni mum t i me, wi t h t he mi ni mum of di sr upt i on t o t he
oper at i on r equi r i ng t he mi ni mum of suppor t and expendi t ur e t hen so much
t he bet t er . As t he consumer ’ s awar eness of , and demand f or , qual i t y,
r el i abi li t y and, avai l abi l i t y i ncr eases, so t oo does t he pr essur e on i ndust r y t o
pr oduce pr oduct s, whi ch meet t hese d emands. I ndust r i es, over t h e year s,
have pl aced gr eat i mpor t ance on engi n eer i ng excel l ence, al t hough some
mi ght pr ef er t o use t he w or d “ hubr i s” . M any of t hose w hi ch have sur vi ved,
however , have done so by manuf act ur i n g hi ghl y r el i abl e pr oduct s, dr i ven by
t he mar ket and t he exp ect at i ons of t hei r cust omer s.
The oper at i onal phase of compl ex equi pm ent l i ke ai r cr af t , r ocket s,
nucl ear submar i nes, t r ai ns, buses, car s and comput er s i s l i ke an or chest r a,
many i ndi vi dual s, i n many depar t ment s d oi ng a set of i nt er connect ed
act i vi t i es t o achi eve maxi mum ef f ect i veness. Behi nd al l of t hese oper at i ons
ar e cer t ai n i nher ent char act er i st i cs (desi gn par amet er s) of t he pr oduct t hat
pl ays a cr uci al r ol e i n t he over al l success of t he pr oduct . Thr ee such
char act er i st i cs ar e r el i abi l i t y, mai nt ai nabi l i t y and suppor t abi l i t y, t oget her
we cal l t h em RM S. Al l t hese t hr ee char act er i st i cs ar e cr uci al f or any
oper at i on. Bi l l i ons of dol l ar s ar e spent by co mmer ci al and mi l i t ar y oper at or s
ever y year as a di r ect co nsequence of t he unr el i abi l i t y, l ack of
mai nt ai nabi l i t y and poor suppor t abi l i t y of t he syst ems t hey ar e expect ed t o
oper at e.
M oder n i ndust r i al syst ems consi st of compl ex and hi ghl y
sophi st i cat ed el ement s, but at t he sam e t i me, user s’ expect at i ons r egar di ng
t r oubl e f r ee oper at i on i s ever pr esent and even i ncr easi n g. A Bo ei ng 777
has over 300, 000 uni que par t s wi t hi n a t ot al of ar ound 6 mi l l i on par t s (hal f
of t hem ar e nu t s, bol t s and r i vet s). Successf ul l y oper at i ng, mai nt ai ni ng and
suppor t i ng such a compl ex syst em demands i nt egr at ed t ool s, pr ocedur es
and t echni ques. Fai l ur e t o meet hi gh r el i abi l i t y, mai nt ai nabi l i t y and
suppor t abi l i t y can have cost l y and f ar r eachi n g ef f ect s. Losi ng t he ser vi ces
1. Rel i abi l i t y Mai n t enance and Logi st i c Suppor t  I nt r oduct i on 4
of ai r l i ner s, such as t he Boei ng 747, can cost as hi gh as $ 300,000 per day i n
f or f ei t ed r evenue al one. Fai l ur e t o di spat ch a commer ci al f l i ght on t i me or
i t s cancel l at i on i s not onl y connect ed t o t h e cost of cor r ect i ng t h e f ai l ur e,
but al so t o t he ext r a cr ew cost s, addi t i onal passenger handl i ng and l oss of
passenger r evenue. Consequ ent l y, t hi s wi l l have an i mpact on t he
comp et i t i veness, pr of i t abi l i t y and mar ket shar e of t he ai r l i ne concer ned.
' Ai r cr af t on Gr ound' i s pr obabl y t h e most dr ead ed phr ase i n t he commer ci al
ai r l i nes’ vocabul ar y. And, al t hough t he cost s and i mpl i cat i ons may be
di f f er en t , i t i s no mor e popul ar wi t h mi l i t ar y oper at o r s. Cost s per mi nut e
del ay f or di f f er ent ai r cr af t t yp e ar e shown i n Fi gur e 1.1. Her e t he del ay
cost s ar e at t r i but abl e t o l abour char ges, ai r por t f ees, ai r t r af f i c cont r ol
cost s, r esch edul i ng cost s, passenger cost s (f ood, acco mmodat i on, t r anspor t
and payof f s).
Fi gur e 1.1 Ai r cr af t del ay cost per mi nut e
Indust r i es have l ear ned f r om past exp er i ence and t hr ough cut t i ng
edge r esear ch how t o mak e t hei r pr odu ct s saf e and r el i abl e. NASA, Boei ng,
Ai r bus, Lockheed M ar t i n, Rol l sRoyce, Gener al El ect r i c, Pr at t and Whi t ney,
and many, many mor e, ar e pr oduci ng ext r emel y r el i abl e pr oduct s. For
exampl e, over 25% of t h e j et l i ner s i n US have been i n ser vi ce f or over 20
year s and mor e t han 500 over 25 year s, near i n g or exceedi ng t hei r or i gi nal
desi gn l i f e (Lam, M ., 1995). The i mpor t an t message i s t hat t hese ai r cr af t ar e
st i l l capabl e of mai nt ai ni ng t hei r ai r w or t hi ness; t hey ar e st i l l saf e and
r el i abl e. But , we cannot b e compl acent , even t he b est of or gani sat i ons can
have t hei r bad days. The l osses of t he Chal l en ger Space Shut t l e i n 1986, and
Apol l o 13 ar e st i l l ver y f r esh i n many of our memor i es.
Cust omer s’ r equi r ement s gener al l y exceed t h e capabi l i t i es of t he
pr oducer s. Occasi onal l y, t hese go beyond what i s pr act i cal l y, and
somet i m es even t heor et i cal l y, possi bl e. An exampl e of t hi s coul d be t he
1. Rel i abi l i t y Mai n t enance and Logi st i c Suppor t  I nt r oduct i on 5
new r el i abi l i t y r equi r ement , mai nt enance and f ai l ur e f r ee oper at i ng per i od,
(Hockl ey et al 1996, Di nesh Kumar et al , 1999, 2000). Hi gh r el i abi l i t y i s
cer t ai nl y a desi r abl e f unct i on, but so t o i s mai nt ai nabi l i t y and excel l ent
l ogi st i c suppor t . It i s onl y t hr ough al l t hr ee t hat t he l i f ecycl e cost can be
dr i ven down whi l st t he l evel of avai l abi l i t y i s dri ven up.
Combat ai r cr af t ar e exp ensi ve and so ar e t hei r cr ews, so no
oper at or want s t o l ose ei t her . At t he same t i me, d epl oyi ng l ar ge gr ou nd
f or ces t o mai nt ai n and suppor t t hem i s al so expen si ve and, pot ent i al l y
hazar dous. It i s t her ef or e not sur p r i si ng t hat t h e oper at or s ar e l ooki ng t o
t he manuf act ur er s t o pr oduce ai r cr af t so r el i abl e t hat t hey can go f or weeks
wi t hout any mai nt enance. The qu est i on i s, however , can we achi eve t he
necessar y l evel of r el i abi l i t y, wi t h suf f i ci ent conf i d ence, at an af f or dabl e
pr i ce, t o meet t hi s r equi r emen t ?.
Recent pr oj ect s such as t he Ul t r a Rel i abl e Ai r cr af t (URA) and Fut ur e
Of f ensi ve Ai r Syst ems (FOAS) pl ace a new di m ensi on t o t he r el i abi l i t y
r equi r ement . The op er at or s/ user s woul d l i ke t o have M ai nt enance Fr ee
Oper at i n g Per i ods (M FOP), dur i ng w hi ch t he pr obabi l i t y t hat t he syst em w i l l
need r est or at i ve mai nt enance i s ver y l ow . Bet ween each of t hese per i ods,
suf f i ci ent mai nt enance wi l l done t o ensur e t he syst em wi l l sur vi ve t h e next
M FOP wi t h t h e same pr obabi l i t y. Onl y t i me wi l l t el l whet her t hi s pol i cy
becomes adopt ed but , t her e i s no doubt t hat t he days of t he M TBF (mean
t i me bet ween f ai l ur es) and i t s i nver se, t he [ const an t ] f ai l ur e r at e ar e sur el y
number ed. Sci ence, mat hemat i cs and pr obabi l i t y t h eor y ar e sl owl y f i ndi ng
t hei r w ay i nt o t he af t er mar ket busi ness and w i t h t h em w i l l come t he need
f or bet t er educat ed peopl e who under st and t hese new concept s,
t echni ques and met hodol ogi es. And, i t wi l l not j ust af f ect mi l i t ar y ai r cr af t ,
buyer s of al l manuf act ur ed pr oduct s wi l l demand gr eat er val ue f or money,
at t he t i me of pur chase, of cour se, but mor e t han t h at t hey wi l l expect i t
t hr oughout i t s l i f e. M anuf act ur er s who have r el i ed on unr el i abi l i t y wi l l
need t o r et hi nk t hei r pol i ci es, pr ocesses and f i nances.
1.2. THE LIFE CYCLE OF ASYSTEM
Fundament al t o any engi neer i ng d esi gn pr act i ce i s an under st andi n g of t he
cycl e, whi ch t he pr oduct goes t hr ou gh dur i ng i t s l i f e. The l i f e cycl e begi ns at
t he moment when an i dea of a new syst em i s bor n and f i ni shes when t he
syst em i s saf el y di sposed. In ot her w or ds, t he l i f e cycl e begi ns w i t h t he
i ni t i al i dent i f i cat i on of t h e needs and r equi r ement s and ext ends t hr ough
pl anni ng, r esear ch, desi gn, pr oduct i on, eval uat i on, oper at i on, mai nt enance,
suppor t and i t s ul t i mat e phase out (Fi gur e 1.2).
1. Rel i abi l i t y Mai n t enance and Logi st i c Suppor t  I nt r oduct i on 6
Fi gur e 1.2 Li f e cycl e of t h e syst em.
M anuf act ur er s w ho speci al i se i n mi l i t ar y har dw ar e w i l l of t en be
appr oached, ei t her di r ect l y or t hr ou gh an adver t i sed “ i nvi t at i on t o t ender ”
t o di scuss t he l at est d ef ence r equi r ement . For most ot her manuf act ur er s, i t
i s gener al l y up t o t hem t o i dent i f y a (pot ent i al ) mar ket need and deci de
whet h er t hey can meet t hat n eed i n a pr of i t abl e way. Th e UK M oD
appr oached BAE Syst ems t o br i ng t oget her a consor t i um (i ncl udi ng
r epr esen t at i ves of t he M oD and RAF) f or an ai r syst em t hat woul d out 
per f or m al l exi st i ng of f ensi ve syst ems, bot h f r i end and f oe, and t hat woul d
i ncl ude al l of t he con cep t s i dent i f i ed as pr act i cal i n t he URA r esear ch
pr oj ect . Ai r bus I ndust r i es, on t he ot her hand, deci ded, based on t h ei r
ext ensi ve mar ket r esear ch, t hat t her e was a suf f i ci ent mar ket n eed f or a
ver y l ar ge ai r cr af t t hat coul d car r y w el l i n excess of 500 passenger s, at l east
acr oss t he Paci f i c f r om Tokyo t o Los Angel es and po ssi bl y even nonst op
bet w een Lon don and Sydney. It wi l l be man y year s bef or e w e wi l l know
whet h er ei t h er of t hese ai r cr af t wi l l get of f t he gr ound and ver y much
l onger t o see i f t hey pr ove a busi ness success f or t hei r manuf act ur er s.
The f i r st pr ocess t hen i s a set of t asks per f or med t o i dent i f y t he
needs and r equi r ement s f or a n ew syst em and t r ansf or m t hem i nt o i t s
t echni cal l y meani n gf ul def i ni t i on. The mai n r easo n f or t he need of a new
syst em coul d be a new f unct i on t o be per f or med (t hat i s t her e i s a new
mar ket d emand f or a pr oduct w i t h t h e speci f i ed f unct i on) or a d ef i ci ency of
t he pr esent syst em. The def i ci enci es coul d be i n t he f or m of : 1. Funct i onal
def i ci en ci es, 2. I nadeq uat e per f or man ce, 3. Inadequat e at t r i but es. 4. Poor
r el i abi li t y, 5. Hi gh mai nt enance and suppor t cost s, 5. Low sal es f i gur es and
hence l ow pr of i t s.
The f i r st st ep i n t he concept ual desi gn phase i s t o anal yse t he
f unct i onal need or d ef i ci ency and t r ansl at e i t i nt o a mor e speci f i c set of
qual i t at i ve and quant i t at i ve r equi r emen t s. Thi s anal ysi s woul d t hen l ead t o
concept ual syst em d esi gn al t er nat i ves. The f l ow of t he concept ual syst em
desi gn pr ocess i s i l l ust r at ed i n Fi gur e 1.3 (D Ver ma and J Knezevi c, 1995).
The out put f r om t hi s st age i s f ed t o t he pr el i mi nar y desi gn st age. The
Needs and
Requirements
Design
Conceptual design
Preliminary design
Detailed design
Production or
Construction
Manufacture
Assembly
Use
Operation
Maintenance
Support
Retirement
1. Rel i abi l i t y Mai n t enance and Logi st i c Suppor t  I nt r oduct i on 7
concept ual desi gn st age i s t he best t i me f or i ncor por at i ng r el i abi l i t y,
mai nt ai nabi l i t y and suppor t abi l i t y consi d er at i ons. In t he case of FOAS, f or
exampl e, var i ous i nt egr at ed pr oj ect t eams wi t h r epr esent at i ves of t he
user s, suppl i er s and even academi a w i l l dr aw n t oget her t o come up w i t h
new i deas and set t ar get s, however , i mpr act i cal . It was l ar gel y a r esul t of
t hi s act i vi t y t hat t h e concept s of t he M FOP and t he u ni nhabi t ed combat ai r
vehi cl e (UCAV) wer e bor n.
Fi gur e 1.3 Concept ual syst em desi gn pr ocess
The mai n t asks dur i ng t he pr el i mi nar y desi gn st age ar e syst em
f unct i onal anal ysi s such as op er at i onal f unct i ons, mai nt enance f u nct i ons,
al l ocat i ons of per f or mance and ef f ect i ven ess f act or s and t he al l ocat i on of
syst em suppor t r equi r ement (Bl anchar d, 1991). It i s at t hi s t i me t hat t he
concept s ar e br ought down t o ear t h ou t of t he “ bl ue sky” . Gr oups wi l l be
r equi r ed t o put t hese i d eal s i nt o r eal i t y possi bl y vi a t echni cal devel opment
pr ogr ams or abandon t hem unt i l t he next t i me.
The mai n t asks per f or med dur i ng t h e det ai l ed desi gn st age 1.
Devel opment of syst em/ pr oduct desi gn, 2. Devel opment of syst em
pr ot ot ype, and 3. Syst em pr ot ot ype t est and eval uat i on. Desi gn i s t h e most
i mpor t ant and cr u ci al st age i n t he pr oduct l i f e cycl e. Rel i abi l i t y,
mai nt ai nabi l i t y and suppor t abi l i t y depend on t he desi gn and ar e t h e mai n
dr i ver s of t he oper at i onal avai l abi l i t y and cost s. It i s dur i ng t hi s st age t hat
saf et y, r el i abi l i t y and mai n t ai nabi l i t y demonst r at i ons can b e per f or med
and, f r om t hese, mai nt enance and suppor t pl ans can be deci d ed.
The pr oduct i on/ const r uct i on pr o cess i s a set of t asks per f or med i n or der
t o t r ansf or m t h e f ul l t echni cal d ef ini t i on of t he n ew syst em i nt o i t s physi cal
exi st en ce. Th e mai n t asks per f or med dur i ng t hi s pr ocess ar e 1.
M anuf act ur e/ Pr odu ct i on/ Test of pr i m e syst em el ement s, 2. Syst em
assessment , 3. Qual i t y Assur ance, and 4. Syst em M odi f i cat i on. Dur i ng t he
pr oduct i on / const r u ct i on pr o cess t he syst em i s physi cal l y cr eat ed i n
accor dance wi t h t he desi gn def i ni t i on. The i nput char act er i st i cs of t he
pr oduct i on pr o cess ar e t he r aw mat er i al , ener gy, equ i pment , f aci l i t i es and
ot her i ngr edi ent s needed f or t h e pr oduct i on/ const r uct i on of t he new
Needs and
Requirements
Needs Analysis
& Requirements
Definition
Synthesis of
conceptual system
design alternatives
Analysis of
Conceptual
System Design
Alternatives
Evaluation of
Conceptual
System Design
Alternatives
1. Rel i abi l i t y Mai n t enance and Logi st i c Suppor t  I nt r oduct i on 8
syst em. The out put char act er i st i cs ar e t he f ul l physi cal exi st ence of t he
f unct i onal syst em.
1.3. CONCEPT OF FAILURE
As w i t h so many w or ds i n t he Engl i sh l anguage, f ai l ur e has come t o
mean man y t hi ngs t o many p eopl e. Essent i al l y, a f ai lur e of a syst em i s any
event or col l ect i on of event s t hat causes t h e syst em t o l ose i t s
f unct i onabi l i t y w her e f unct i onabi l i t y i s t he i nher ent char act er i st i c of a
pr oduct r el at ed t o i t s abi l i t y t o per f or m a speci f i ed f unct i on accor di ng t o t he
speci f i ed r equi r ement s under t he speci f i ed oper at i ng condi t i ons. ( Kn ezevi c
1993) Thus a syst em, or i ndeed , any compon ent wi t hi n i t , can onl y be i n
one of t wo st at es: st at e of f unct i oni ng or ; st at e of f ai l ur e.
In many cases, t h e t r ansi t i on bet ween t hese st at es i s ef f ect i vel y
i nst ant aneous; a wi ndscr een shat t er s, a t yr e punct ur es, a bl ade br eaks, a
t r ansi st or bl ow s. Ther e i s i nsuf f i ci ent t i me t o det ect t he onset or pr even t
t he consequences. However , i n many ot her cases, t he t r ansi t i on i s gr adual ;
a t yr e or bear i ng wear s, a cr ack pr opagat es acr oss a di sc, a bl ade “ cr eeps”
or t he p er f or mance st ar t s t o dr op of f . In t hese ci r cumst ances, some f or m
of heal t h moni t or i ng may al l ow t he user t o t ake pr even t at i ve measur es.
Inspect i ng t he amount of t r ead on t he t yr es at r egul ar i nt er val s, scanni ng
t he l ubr i cat i ng oi l f or excessi ve d ebr i s, bor oscope i nspect i on t o l ook f or
cr acks or usi ng some f or m t r endi ng (e.g. Kal man Fi l t er i ng) on t he speci f i c
f uel consumpt i on can al er t t h e user t o i mmi n ent onset of f ai l ur e. Si mi l arl y,
any one of t he many f or ms of nond est r u ct i ve t est i ng may be used (as
appr opr i at e) on compon ent s t hat have been exposed dur i ng t he r ecover y
of t hei r par ent component t o ch eck f or damage, det er i or at i on, er osi on,
cor r osi on or any of t he ot h er vi si bl e or physi cal l y det ect abl e si gns t hat
mi ght cause t he co mponent t o become nonf unct i onabl e.
Wi t h man y hi ghl y co mpl ex syst ems, whose f ai l ur e may have ser i ous
or cat ast r ophi c consequences, measur es ar e t aken, wher ever possi bl e, t o
mi t i gat e agai nst such event s. Car s ar e f i t t ed w i t h dual br aki ng syst ems,
ai r cr af t wi t h (at l east ) t r i pl e hydr aul i c syst ems and numer ous ot her
i nst ances of r edundancy. I n t hese cases, i t i s possi bl e t o have a f ai l ur e of a
compon en t wi t hout a f ai l ur e of t he syst em. The r ecover y of t he f ai l ed i t em,
vi a a mai nt enance act i on, may be def er r ed t o a t i me whi ch i s mor e
conveni ent t o t he oper at or , saf e i n t he knowl edge t hat t her e i s an
accept abl y hi gh pr obabi l i t y t hat t he syst em wi l l cont i nue oper at i ng saf el y
f or a cer t ai n l engt h of t i me. If one of t he f l i ght cont r ol comput er s on an
ai r cr af t f ai l s, i t s f unct i ons wi l l i nst ant l y and aut omat i cal l y be t aken over by
1. Rel i abi l i t y Mai n t enance and Logi st i c Suppor t  I nt r oduct i on 9
one of t h e ot her comput er s. The f l i ght wi l l gener al l y be al l owed t o
cont i nue, uni nt er r u pt ed t o i t s next schedul ed dest i nat i on. Dependi ng on
t he l evel of r edundancy and r egul at i ons/ cer t i f i cat i on, f ur t her f l i ght s may be
per mi t t ed, ei t her unt i l anot her comput er f ai l s or , t he ai r cr af t i s put i n f or
schedul ed mai nt enance.
M ost co mmer ci al ai r li ner s ar e f i t t ed w i t h t w o, or mor e, engi nes. Par t of t he
cer t i f i cat i on pr ocess r equi r es a pr act i cal demonst r at i on t hat a f ul l y l oaded
ai r cr af t can t akeof f saf el y even i f one of t hose engi nes f ai l s at t he most
cr i t i cal t i me; “ r ot at i on” or “ wei ght of f wheel s” . However , even t hough t he
ai r cr af t can f l y wi t h one engi ne out of ser vi ce, once i t has l anded, i t woul d
not t hen be p er mi t t ed t o t akeof f agai n unt i l t hat en gine has been r et ur ned
t o a st at e of f unct i oni ng (except und er ver y excep t i onal ci r cumst ances) .
Wi t h t he l at est l ar ge t w i ns (e.g. Ai r bus 330 and Bo ei ng 777), a change i n t he
ai r wor t hi ness r ul es has al l owed t hem t o f l y f or ext ended per i ods f ol l owi ng
t he i nf l i ght shut down of one of t he engi n es, gen er al l y r ef er r ed t o ETOPS
(whi ch of f i ci al l y st ands f or ext ended t wi n oper at i ons over sea or ,
unof f i ci al l y, engi nes t ur n or passenger s sw i m). Thi s def i nes t he maxi mum
di st ance (usual l y expr essed i n mi nut es of f l yi ng t i me) t he ai r cr af t can be
f r om a sui t abl e l andi ng si t e at any t i me dur i ng t h e f l i ght . It al so r equi r es an
ai r cr af t t hat has “ l ost ” an en gi ne t o f l y t o i m medi at el y di ver t t o a l andi ng
si t e t hat i s w i t hi n t hi s f l yi ng t i me. Agai n, havi ng l anded, t hat ai r cr af t w ould
not be per mi t t ed t o t ake of f unt i l i t was f i t t ed wi t h t wo f unct i onabl e
engi n es. In t hi s case, nei t her engi ne i s t r ul y r edundant but , t he syst em
(ai r cr af t ) has a l i mi t ed l evel of f aul t / f ai l ur e t ol er ance.
M ost per sonal comput er s (PC) come compl et e wi t h a “ har d di sc” .
Dur i ng t he l i f e of t he PC, i t i s not uncommon f or smal l sect or s of t hese di scs
t o become unusabl e. Pr ovi ded t h e sect or di d not hol d t he f i l e access t abl e
(FAT) or key syst em’ s f i l es, t he comput er i s not onl y abl e t o d et ect t h ese
sect or s but i t w i l l mar k t hem as unusabl e and avoi d w r i t i ng any dat a t o
t hem. Unf or t unat el y, i f t her e was al r eady dat a on t hese sect or s bef or e
t hey become unusabl e, t hi s w i l l no l onger be accessi bl e, al t hough w i t h
speci al sof t w ar e, i t may be possi bl e t o r ecover some o f i t . Thus, t he bui l t i n
t est sof t war e of t he comput er i s abl e t o pr ovi de a l evel of f aul t t ol er ance
whi ch i s of t en t ot al l y i nvi si bl e t o t he user , at l east unt i l t he whol e di sc
cr ashes or t he f aul t af f ect s a cr i t i cal par t of a pr ogr am or dat a. Even under
t hese ci r cumst ances, i f t hat pr o gr am or dat a has been backed up t o ano t her
medi u m, i t shoul d be possi bl e t o r est or e t he f ul l capaci t y of t he syst em
usual l y wi t h a l evel of manual i nt er vent i on. So t her e i s bot h f aul t t ol er ance
and r edundancy al t hough t he l at t er i s usual l y at t he di scr et i on of t he user .
2. 10
Chapter 2
Probability Theory
We do not know how t o pr edi ct what woul d happen i n any gi ven
ci r cumst ances, and we bel i eve now t hat i t i s possi bl e, t hat t he onl y t hi ng t hat
can be pr edi ct ed i s t he pr obabi l i t y of di f f er ent event s
Ri char d Feynman
Pr obabi l i t y t heor y pl ays a l eadi ng r ol e i n moder n sci en ce i n spi t e of t he f act
t hat i t w as i ni t i al l y devel oped as a t ool t hat coul d be used f or guessi ng t he
out come of some games of chance. Pr obabi l i t y t heor y i s appl i cabl e t o
ever yday l i f e si t uat i ons wher e t he out co me of a r epeat ed pr ocess,
exp er i ment , t est , or t r i al i s uncer t ai n and a pr edi ct i on has t o be made.
In or der t o appl y pr obabi l i t y t o ever yday engi neer i ng p r act i ce i t i s necessar y
t o l ear n t he t er mi nol ogy, def i ni t i ons and r ul es of pr obabi l i t y t heor y. Thi s
chapt er i s not i nt ended t o a r i gor o us t r eat men t of al l r el evant t heor ems
and pr oof s. The i nt ent i on i s t o pr ovi de an under st andi ng of t he mai n
concept s i n pr obabi l i t y t heor y t hat can be appl i ed t o pr obl ems i n r el i abi l i t y,
mai nt enan ce and l ogi st i c suppor t , whi ch ar e di scussed i n t he f ol l owi ng
chapt er s.
2.4. PROBABILITY TERMS AND DEFINITIONS
In t hi s sect i on t hose el ement s essen t i al f or under st andi ng t he r udi ment s of
el ement ar y pr obabi l i t y t heor y w i l l be di scussed and def i ned i n a gener al
2. Pr obabi l i t y Theor y 11
manner , t oget her wi t h i l l ust r at i ve exampl es r el at ed t o engi neer i ng pr act i ce.
To f aci l i t at e t he di scussi on some r el evant t er ms and t hei r def i ni t i ons ar e
i nt r oduced.
Experiment
An exper i men t i s a wel l def i ned act or pr ocess t hat l eads t o a si ngl e wel l 
def i ned out com e. Fi gur e 2.1 i l l ust r at es t h e con cept of r andom exper i ment s.
Ever y exper i ment must :
1. Be capable of being described, so that the observer knows when it occurs.
2. Have one and only one outcome, so that the set of all possible outcomes
can be specified.
Fi gur e 2.1 Gr aphi cal Repr esent at i on of an Exper i ment and i t s out comes.
Elementary event
An elementary event is every separate outcome of an experiment.
From the definition of an experiment, it is possible to conclude that
the total number of elementary events is equal to the total number of
possible outcomes, since every experiment must have only one
outcome.
Sample space
The set of all possible distinct outcomes for an experiment is called
the sample space for that experiment.
M ost f r equent l y i n t he l i t er at ur e t h e symbol S i s used t o r epr esent t he
sampl e space, and smal l l et t er s, a, b, c, .., f or el ement ar y event s t hat ar e
possi bl e out comes of t he exper i men t und er consi der at i on. The set S may
Experiment
2. Pr obabi l i t y Theor y 12
cont ai n ei t her a f i ni t e or an i nf i n i t e number of el ement ar y event s. Fi gur e
2.2 i s a gr aphi cal pr esent at i on of t he sampl e space.
Fi gur e 2.2 Gr aphi cal Pr esent at i on of t he Sampl e Space
Event
Event is a subset of the sample space, that is, a collection of
elementary events.
Capi t al l et t er s A, B, C, …, ar e usual l y used f or denot i ng event s. For exampl e,
i f t he exp er i ment per f or med i s measur i ng t he sp eed of passi ng car s at a
speci f i c r oad j unct i on, t hen t he el ement ar y event i s t he sp eed measur ed,
wher eas t he sampl e space consi st s of al l t he di f f er ent speeds one mi ght
possi bl y r ecor d. Al l speed event s co ul d be cl assi f i ed i n, say, f our di f f er ent
speed gr oups: A (l ess t han 30 km/ h), B (bet ween 30 and 50 km/ h), C
(bet w een 50 and 70 km/ h) and D (above 70 km/ h) . I f t he measur ed sp eed
of t he passi ng car i s, say 35 km/ h, t hen t he event B i s sai d t o have occur r ed.
2.5. ELEMENTARY THEORY OF PROBABILITY
The t heor y of pr obabi l i t y i s devel oped f r om axi oms pr oposed by t he
Russi an mat h emat i ci an Kol mogr ov. In pr act i ce t hi s means t hat i t s el ement s
have been def i ned t oget h er w i t h sever al axi oms w h i ch go ver n t h ei r
r el at i ons. Al l ot her r ul es and r el at i ons ar e der i ved f r om t hem.
2.5.1 Axioms of Probability
In cases w her e t he o ut come of an exper i ment i s uncer t ai n, i t i s
necessar y t o assi gn some m easur e t hat w i l l indi cat e t h e chances of
2. Pr obabi l i t y Theor y 13
occur r ence o f a p ar t i cul ar even t . Such a measur e o f event s i s cal l ed
t he pr obabi l i t y of t he event an d symbol i sed b y P(.), ( P(A) d enot es t he
pr obabi l i t y of even t A). The f un ct i on whi ch associ at es each event A in
t he sampl e space S, wi t h t he pr ob ab i l i t y measur e P(A), i s cal l ed t he
pr obabi l i t y f unct ion  t he pr obabi l i t y of t hat event . A gr aph i cal
r ep r esent at i on of t he pr obabi l i t y f unct i on i s gi ven i n Fi gu re 2.3.
Fi gur e 2.3 Gr aphi cal r epr esent at i on o f p r obabi l i t y f u nct i on.
Formally, the probability function is defined as:
A function which associates with each event A, a real number, P(A),
the probability of event A, such that the following axioms are true:
1. P(A) > 0 for every event A,
2. P(S) = 1, (probability of the sample space)
3. The probability of the union of mutually exclusive events is the sum of
their probabilities, that is
) ( ... ) ( ) ( ) ... (
2 1 2 1 n n
A P A P A P A A A P + + + · ∪ ∪
In essence, t hi s d ef i ni t i on st at es t hat each event A i s pai r ed wi t h a non
negat i ve number , pr obabi l i t y P(A), and t hat t he pr obabi l i t y of t he sur e
event S, or P(S), i s al w ays 1.
Furthermore, if A
1
and A
2
are any two mutually exclusive events (that is,
the occurrence of one event implies the nonoccurrence of the other) in the
2. Pr obabi l i t y Theor y 14
sample space, the probability of their union P A A ( )
1 2
∪ , is simply the sum
of their two probabilities, P A P A ( ) ( )
1 2
+ .
2.5.2 Rules of Probability
The f ol l owi ng el ement ar y r ul es of pr obabi l i t y ar e di r ect l y deduced f r om t he
or i gi nal t hr ee axi oms, usi ng t he set t heor y:
a) For any event A, the probability of the complementary event, written A' ,
is given by
P A P A ( ' ) ( ) · − 1 (2.1)
b) The probability of any event must lie between zero and one inclusive:
0 1 ≤ ≤ P A ( ) (2.2)
c) The probability of an empty or impossible event, φ, is zero.
P( ) φ · 0 (2.3)
d) If occurrence of an event A implies that an event B occurs, so that the
event class A is a subset of event class B, then the probability of A is less
than or equal to the probability of B:
) ( ) ( B P A P ≤ (2.4)
e) In order to find the probability that A or B or both occur, the probability
of A, the probability of B, and also the probability that both occur must
be known, thus:
P A B P A P B P A B ( ) ( ) ( ) ( ) ∪ · + − ∩ (2.5)
f) If A and B are mutually exclusive events, so that P A B ( ) ∩ · 0, then
P A B P A P B ( ) ( ) ( ) ∪ · + (2.6)
g) If n events form a partition of S, then their probabilities must add up to
one:
2. Pr obabi l i t y Theor y 15
∑ · · + + +
·
n
i
i n
A P A P A P A P
1
2 1
1 ) ( ) ( ... ) ( ) ( (2.7)
2.5.3 Joint Events
Any event that is an intersection of two or more events is a joint event.
Ther e i s not hi ng t o r est r i ct any gi ven el ement ar y event f r om t he sampl e
space f r om qual i f yi ng f or t wo or mor e event s, pr ovi ded t hat t hose event s
ar e not mu t ual l y excl usi ve. Thus, gi ven t he event A and t he event B, t he
j oi nt event i s A B ∩ . Si nce a member of A B ∩ must be a member of set
A, and al so of set B, bot h A and B event s o ccur when A B ∩ occur s.
Pr ovi ded t hat t he el ement s of set S ar e al l equal l y l i kel y t o occur , t he
pr obabi l i t y of t he j oi nt even t coul d be f ound i n t he f ol l owi ng way:
P A B ( ) ∩ ·
∩ number of elementary events in A B
total number of elementary events
2.5.4 Conditional Probability
If A and B ar e event s i n a sampl e space w hi ch consi st s of a f i ni t e number of
el ement ar y even t s, t he condi t i onal pr obabi l i t y of t he event B gi ven t hat t he
event A has al r eady occur r ed, denot ed by P B A (  ) , i s def i ned as:
P B A P A (  ) , ( ) ·
∩
>
P(A B)
P(A)
0 (2.8)
2. Pr obabi l i t y Theor y 16
Fi gur e 2.4 Gr aphi cal Pr esent at i on of t he Bayes Theor em
The condi t i onal pr obabi l i t y symbol , P B A (  ) , i s r ead as t he pr obabi l i t y of B
gi ven A. It i s necessar y t o sat i sf y t he condi t i on t hat P(A)>0, because i t does
not make sense t o consi d er t he pr obabi l i t y of B gi ven A i f event A i s
i mpossi bl e. For any t wo event s A and B, t her e ar e t w o condi t i onal
pr obabi l i t i es t hat may be cal cul at ed:
P B A and P A B (  ) (  ) ·
∩
·
∩ P(A B)
P(A)
P(A B)
P(B)
(The probability of B, given A) (The probability of A,
given B)
One of t h e i mpor t ant appl i cat i on of condi t i onal pr obabi l i t y i s due t o Bayes
t heor em, whi ch can be st at ed as f ol l ows:
If ( , , , ) A A A
N 1 2
K r epr esen t s t h e par t i t i on of t he sampl e space (N
mut ual l y excl usi ve event s), and i f B i s subset of ( ) A A A
N 1 2
∪ ∪ ∪ K , as
i l l ust r at ed i n Fi gur e 2.4, t hen
P A B
i
(  )
)
) ) )
·
+ + + +
P(BA )P(A
P(BA )P(A P(BA )P(A P(BA )P(A
i i
1 1 i i N N
K K
(2.9)
2.6. PROBABILITY AND EXPERIMENTAL DATA
The cl assi cal appr oach t o pr obabi l i t y est i mat i on i s based on t he r el at i ve
f r equency of t he o ccur r ence of t hat event . A st at ement of pr obabi l i t y t el l s
us what t o expect about t h e r el at i ve f r equency of occur r ence, gi ven t hat
enough obser vat i ons ar e mad e. In t he l ong r un, t he r el at i ve f r eq uency of
occur r ence of an event , say A, shoul d appr oach t he pr obabi l i t y of t hi s
event , i f i ndependent t r i al s ar e made at r andom over an i ndef i ni t el y l ong
sequ ence. Thi s pr i nci pl e was f i r st f or mul at ed an d pr oved b y James
Ber noul l i i n t he ear l y ei ght eent h cent ur y, and i s now wel l known as
Ber noul l i ' s t heor em:
If t he pr obabi l i t y of occur r ence of an event A i s p, and i f n t r i al s ar e made
i ndepen dent l y and und er t he same condi t i ons, t hen t he pr obabi l i t y t hat t he
r el at i ve f r equ ency of occur r ence of A, (def i ned as f A N A n ( ) ( ) · ) di f f er s
2. Pr obabi l i t y Theor y 17
f r om p by an y amount , how ever smal l , appr oaches zer o as t he number of
t r i al s gr ows i ndef i ni t el y l ar ge. That i s,
P N A n p s as n ( ( ) )  ) , − > → → ∞ 0 (2.10)
where s is some arbitrarily small positive number. This does not mean that
the proportion of
n
A N ) (
occurrences among any n trial must be p; the
proportion actually observed might be any number between 0 and 1.
Nevertheless, given more and more trials, the relative frequency of f A ( )
occurrences may be expected to become closer and closer to p.
Al t hough it i s t r ue t hat t he r elat i ve f r equ en cy of occur r ence of an y event i s
exact l y equal t o t he pr obabi l i t y of occur r ence of any event onl y f or an
i nf i ni t e numb er of i ndepend ent t r i al s, t hi s poi nt must not b e over st r essed.
Even wi t h r el at i vel y smal l number of t r i al s, t her e i s ver y good r eason t o
exp ect t he obser ved r el at i ve f r equ en cy t o b e qui t e cl ose t o t he pr obabi l i t y
because t he r at e of con ver gence of t he t wo i s ver y r api d. How ever , t he
mai n dr aw back of t he r el at i ve f r equency appr oach i s t hat i t assumes t hat al l
event s ar e equal l y l i kel y (equal l y pr obabl e).
2.7. PROBABILITY DISTRIBUTION
Consider the set of events A A A
n 1 2
, , , K , and suppose that they form a
partition of the sample space S. That is, they are mutually exclusive and
exhaustive. The corresponding set of probabilities, P A P A P A
n
( ), ( ), , ( )
1 2
K ,
is a probability distribution. An illustrative presentation of the concept of
probability distribution is shown in Figure 2.5.
As a si mpl e exampl e of a pr obabi l i t y di st r i but i on, i magi ne a sampl e space of
al l For d car s pr oduced. A car sel ect ed at r ando m i s cl assi f i ed as a sal oon or
coupe or est at e. The pr obabi l i t y di st r i but i on mi ght be:
Event Saloon Coupe Estate Total
P 0.60 0.31 0.09 1.00
Al l even t s ot her t han t hose l i st ed have pr obabi l i t i es of zer o
2. Pr obabi l i t y Theor y 18
Fi gur e 2.5 Gr aphi cal r epr esent at i on of Pr obabi l i t y Di st r i but i on
2.8. RANDOM VARIABLE
A f unct i on t hat assi gns a number (usual l y a r eal number ) t o each sampl e
poi nt i n t he sampl e space S i s a r andom var i abl e.
Out comes of exp er i ment s may be expr essed i n numer i cal and non
numer i cal t er ms. In or der t o compar e and anal yse t h em i t i s much mor e
conveni ent t o deal wi t h numer i cal t er ms. So, f or pr act i cal appl i cat i ons, i t is
necessar y t o assi gn a numer i cal val ue t o each possi bl e el ement ar y event i n
a sampl e space S. Even i f t he el ement ar y event s t hemsel ves ar e al r eady
expr essed i n t er ms of number s, i t i s possi bl e t o r eassi gn a uni que r eal
number t o each el ement ar y event . The f unct i on t hat achi eves t hi s i s know n
as t he r andom var i abl e. In ot her w or ds, a r andom var i abl e i s a r eal val ued
f unct i on def i ned i n a sampl e space. Usual l y i t i s denot ed wi t h capi t al l et t er s,
such as X, Y and Z, w her eas smal l l et t er s, such as x, y, z, a, b, c, and so on,
ar e used t o denot e par t i cul ar val u es of r andom var i abl es, see Fi gur e 2.6
If X is a random variable and r is a fixed real number, it is possible to
define the event A to be the subset of S consisting of all sample points 'a' to
which the random variable X assigns the number r, ) ) ( : ( r a X a A · · . On
the other hand, the event A has a probability ) ( A P p · . The symbol p can
be interpreted, generally, as the probability that the random variable X takes
on the value r, ) ( r X P p · · . Thus, the symbol ) ( r X P · represents the
probability function of a random variable.
0 p1 p2 ... pi ...pn 1
Sample Space (S)
a1 a2
an
}
Probability Distribution
S
P
a1
p1
a2
p2
.
.
an
pn
2. Pr obabi l i t y Theor y 19
Fi gur e 2.6 Gr aphi cal Repr esent at i on of Random Var i abl e
Ther ef or e, by usi ng t he r andom var i abl e i t i s possi bl e t o assi gn pr obabi l i t i es
t o r eal number s, al t hough t he or i gi nal pr obabi l i t i es wer e o nl y def i ned f or
event s of t he set S, as show n i n Fi gur e 2. 7.
The pr obabi l i t y t hat t he r andom var i abl e X, t akes val u e l ess t han or equal t o
cer t ai n val ue ' x' , is cal led t he cumul at i ve di st r i but i on f unct i on, F(t ). That i s,
P[ X ≤ x] = F(x)
Fi gur e 2.7 Rel at i onshi p bet w een pr obabi l i t y f unct i on and a r andom var i abl e
2.8.1 Types of random variables
Depending on the values, which the random variables can assume,
random variables, can be classified as discrete or continuous. The main
characteristics, similarities and differences for both types will be briefly
described below.
Di scret e random vari abl es
If the random variable X can assume only a particular finite or countably
infinite set of values, it is said to be a discrete random variable.
Ther e ar e ver y many si t uat i ons wher e t he r andom var i abl e X can assume
onl y a par t i cul ar f i ni t e or count abl y i nf i ni t e set of val ues; t hat i s, t he
possi bl e val ues of X ar e f i ni t e i n number or t hey ar e i nf i ni t e i n number but
can be put i n a onet oon e cor r espond ence wi t h a set of r eal number .
2. Pr obabi l i t y Theor y 20
Cont i nuous r andom vari abl es
If the random variable X can assume any value from a finite or an infinite
set of values, it is said to be a continuous random variable.
Let us consi der an exp er i ment , whi ch consi st s of r ecor di ng t he t emper at ur e
of a cool i ng l i qui d of an engi ne i n t h e ar ea of t he t h er most at at a gi ven
t i me. Suppose t hat we can measur e t he t emper at ur e exact l y, whi ch means
t hat our measur i ng devi ce al l ows us t o r ecor d t he t emper at ur e t o any
number of d eci mal poi nt s. If X i s t he t emper at ur e r eadi ng, i t i s not possi bl e
f or us t o sp eci f y a f i ni t e or count abl y i nf i ni t e set of val u es. For exampl e, i f
one of t he f i ni t e set of val u es i s 75.965, we can det er mi ne val ues 75.9651,
75.9652, and so on, w hi ch ar e al so possi bl e val ues of X. What i s bei ng
demonst r at ed her e i s t hat t he possibl e val ues of X consi st of t he set of r eal
number s, a set whi ch cont ai ns an i nf i ni t e (and uncount abl e) number of
val ues.
Cont i nuous r andom var i abl es have enor mous ut i l i t y i n r el i abi l i t y,
mai nt enance and l ogi st i c suppor t as t he r andom var i abl es t i me t o f ai l ur e,
t i me t o r epai r and t he l ogi st i c del ay t i me ar e con t i nuous r andom var i abl es.
2.9. THE PROBABILITY DISTRIBUTION OF
RANDOM VARIABLE
Taking into account the concept of the probability distribution and the
concept of the random variable, it could be said that the probability
distribution of the random variable is a set of pairs, [ ¦ r P X r i n
i i
, ( ), , · · 1
as shown in Figure 2.8.
2. Pr obabi l i t y Theor y 21
Figure 2.8 Probability Distribution of a Random Variable
The easiest way to present this set is to make a list of all its members. If
the number of possible values is small, it is easy to specify a probability
distribution. On the other hand, if there are a large number of possible
values, a listing may become very difficult. In the extreme case where we
have an infinite number of possible values (for example, all real numbers
between zero and one), it is clearly impossible to make a listing.
Fortunately, there are other methods that could be used for specifying a
probability distribution of a random variable:
a) Functional method, where a specific mathematical functions exist from
which the probability of any value or interval of values can be calculated.
b) Parametric method, where the entire distribution is represented through
one or more parameters known as summary measures.
2.9.1 Functional Method
By d ef i ni t i on, a f unct i on i s a r el at i on w her e each m ember of t he domai n i s
pai r ed wi t h on e memb er of t he r ange. I n t hi s par t i cu l ar case, t he r el at i on
bet w een numer i cal val ues w hi ch r andom var i abl es can have and t hei r
pr obabi l i t i es wi l l be consi der ed. Th e most f r equent l y used f unct i ons f or t he
descr i pt i on of pr obabi l i t y di st r i but i on of a r andom var i abl e ar e t he
pr obabi l i t y mass f unct i on, t he pr obabi l i t y densi t y f unct i on, and t he
cumul at i ve di st r i but i on f un ct i on. Each of t hese wi l l be anal ysed and
def i ned i n t he r emai nder of t hi s chapt er .
Probabi l i t y mass f unct i on
2. Pr obabi l i t y Theor y 22
This function is related to a discrete random variable and it represents the
probability that the discrete random variable, X, will take one specific value
x
i
, p P X x
i i
· · ( ) . Thus, a probability mass function, which is usually
denoted as PMF(.) , places a mass of probability p
i
at the point of x
i
on
the Xaxis.
Given that a discrete random variable takes on only n different values, say
a a a
n 1 2
, , , K , the corresponding PMF(.) must satisfy the following two
conditions:
1 0 1 2
2 1
1
. ( ) , , ,
. ( )
P X a for i n
P X a
i
i
i
n
· ≥ ·
· ·
·
∑
K
(2.11)
In practice this means that the probability of each value that X can take
must be nonnegative and the sum of the probabilities must be 1.
Thus, a probability distribution can be represented by the set of pairs
of values ( , ) a p
i i
, where i n · 1 2 , , , K , as shown in Figure 2.9. The
advant age of such a gr aph over a l i st i ng i s t he ease of compr ehensi on and a
bet t er pr ovi si on of a not i on f or t he nat ur e of t he pr ob abi l i t y di st ri bu t i on.
Fi gur e 2.9 Pr obabi l i t y M ass Funct i on
Probabi l i t y densi t y f unct i on
In the previous section, discrete random variables were discussed in terms
of probabilities P(X =x), the probability that the random variables take on an
exact value. However, consider the example of an infinite set for a specific
type of car, where the volume of the fuel in the fuel tank is measured with
only some degree of accuracy. What is the probability that a car selected at
random will have exactly 16 litres of fuel? This could be considered as an
event that is defined by the interval of values between, say 15.5 and 16.5, or
15.75 and 16.25, or any other interval t × 16 01 . i , where i is not exactly
2. Pr obabi l i t y Theor y 23
zero. Since the smaller the interval, the smaller the probability, the
probability of exactly 16 litres is, in effect, zero.
In general, for continuous random variables, the occurrence of any exact
value of X may be regarded as having zero probability.
The Probability Density Function, ) (x f , which represents the probability
that the random variable will take values within the interval
x X x x ≤ ≤ + ∆( ) , when ∆( ) x approaches zero, is defined as:
f x
P x X x x
x x
( ) lim
( ( ))
( )
·
≤ ≤ +
→ ∆
∆
∆ 0
(2.12)
As a consequence, t he pr obabi l i t i es of a cont i nuous r andom var i abl e can be
di scussed onl y f or i nt er val s of X val ues. Thus, i nst ead of t he pr obabi l i t y t hat
X t akes on a sp eci f i c val ue, say ' a' , w e d eal w i t h t he socal l ed pr obabi l i t y
densi t y of X at ' a' , symbol i sed by f a ( ) . I n gener al , t he pr obabi l i t y
di st r i but i on of a cont i nuous r andom var i abl e can be r epr esent ed b y i t s
Pr obabi l i t y Densi t y Funct i on, PDF, w hi ch i s def i ned i n t he f ol l ow i n g w ay:
P a X b f x dx
a
b
( ) ( ) ≤ ≤ ·
∫
(2.13)
A f ul l y def i ned pr obabi l i t y densi t y f unct i on must sat i sf y t h e f ol l owi ng t wo
r equi r ement s:
f x for all x ( ) ≥ 0
f x dx ( )
−∞
+∞
∫
· 1
The PDF i s al w ays r epr esen t ed as a smo ot h cur ve dr aw n abo ve t he
hor i zont al axi s, w hi ch r epr esent s t he possi bl e val ues of t he r andom var i abl e
X. A cur ve f or a hypot het i cal di st r i but i on i s show n i n Fi gur e 2.10 w her e t he
t w o poi nt s a and b o n t he hor i zont al axi s r epr esent l i mi t s whi ch def i n e an
i nt er val .
2. Pr obabi l i t y Theor y 24
Fi gur e 2.10 Pr obabi l i t y Densi t y Funct i on f or a Hypot het i cal Di st r i but i on
The shaded por t i on b et w een ' a' and ' b' r epr esent s t he pr obabi l i t y t hat X
t akes on a val ue bet ween t he l i mi t s ' a' and ' b' .
Cumul at i ve di st ri but i on f unct i on
The pr obabi l i t y t hat a r andom var i abl e X t akes on a val ue at or bel ow a
gi ven number ' a' i s of t en wr i t t en as:
) ( ) ( a X P a F ≤ · (2.14)
The symbol ) (a F denot es t he par t i cul ar pr obabi l i t y f or t he i nt er val a X ≤ .
The gener al symb ol ) (x F i s somet i mes used t o r epr esent t h e f unct i on
r el at i ng t h e var i ous val ues of X t o t h e cor r espondi ng cumul at i ve
pr obabi l i t i es. Thi s f unct i on i s cal l ed t he Cumul at i ve Di st r i but i on Funct i on,
CDF, and i t must sat i sf y cer t ai n mat h emat i cal pr oper t i es, t he most
i mpor t ant of whi ch ar e:
1. 0 1 ≤ ≤ F x ( )
2. if a b F a F b < ≤ , ( ) ( )
3. F and F ( ) ( ) ∞ · −∞ · 1 0
2. Pr obabi l i t y Theor y 25
Figure 2.11 Cumulative Distribution Function for Discrete Variable
Figure 2.12 Cumulative Distribution Function for Continuous Variable
The symbol F x ( ) can be used to represent the cumulative probability that
X is less than or equal to x. It is defined as:
F a P X x
i
i
n
( ) ( ) · ·
·
∑
1
(2.15)
For t he di scr et e r andom var i abl es, wher eas i n t he case of cont i nuous
r andom var i abl es i t wi l l t ake t he f ol l owi ng f or m:
F a f x dx
a
( ) ( ) ·
−∞
∫
(2.16)
Hypot het i cal cu mul at i ve di st r i but i on f unct i ons f or bot h t yp es of r andom
var i abl e ar e gi ven i n Fi gur es 2.11 and 2.12.
2.9.2 Parametric Method
In some si t uat i ons i t i s easi er and even mor e ef f i ci ent t o l ook onl y at cer t ai n
char act er i st i cs of di st r i but i ons r at her t han t o at t empt t o speci f y t he
di st r i but i on as a whol e. Such char act er i st i cs su mmar i se and nu mer i cal l y
descr i b e cer t ai n f eat ur es f or t he ent i r e di st r i but i on. Two gen er al gr oups of
such char act er i st i cs appl i cabl e t o any t ype of di st r i but ion ar e:
a) Measures of central tendency (or location) which indicate the typical or
the average value of the random variable.
0
1
x
F(x)
a
F(a)
2. Pr obabi l i t y Theor y 26
b) Measures of dispersion (or variability) which show the spread of the
difference among the possible values of the random variable.
In many cases, i t i s possi bl e t o adequat el y descr i be a pr obabi l i t y di st r i but i on
wi t h a f ew measur es of t hi s ki nd. It shoul d b e r emember ed, however , t hat
t hese measur es ser ve onl y t o summar i se some i mpo r t ant f eat ur es of t he
pr obabi l i t y di st r i but i on. In gener al , t hey do not co mpl et el y d escr i be t he
ent i r e di st r i but i on.
One of t h e most co mmon and usef ul sum mar y measur es of a pr obabi l i t y
di st r i but i on i s t he expect at i on of a r andom var i abl e, E(X). It i s a uni que
val ue t hat i ndi cat es a l ocat i on f or t he di st r i but i on as a whol e (In physi cal
sci ence, expect ed val u e act ual l y r epr esent s t he Cent r e of gr avi t y). The
concept of expect at i on pl ays an i mpor t ant r ol e n ot onl y as a usef ul
measur e, but al so as a cent r al concept wi t hi n t he t heor y of pr obabi l i t y and
st at i st i cs.
If a r andom var i abl e, say X, i s di scr et e, t hen i t s expect at i on i s def i ned as:
∑
· × ·
x
x X P x X E ) ( ) ( (2.17)
Wh er e t he su m i s t aken f or al l t he val ues t hat t he var i abl e X can assume. If
t he r andom var i abl e i s cont i nuous, t he expect at i on i s def i ned as:
∫
+∞
∞ −
× · dx x f x X E ) ( ) ( (2.18)
Wh er e t he sum i s t aken o ver al l val ues t hat X can assu me. For a cont i nuous
r andom var i abl e t he exp ect at i on i s def i ned as:
E X F x dx ( ) [ ( )] · −
−∞
+∞
∫
1 (2.19)
If c is a constant, then
) ( ) ( X E c cX E × · (2.20)
Al so, f or any t wo r andom var i abl es X and Y,
) ( ) ( ) ( Y E X E Y X E + · +
2. Pr obabi l i t y Theor y 27
Measures of cent ral t endency
The most f r equ ent l y used measur es ar e:
The mean of a r andom var i abl e i s si mpl y t he expect at i on of t he r andom
var i abl e under consi der at i on. Thus, f or t he r andom var i abl e, X, t he mean
val ue i s def i ned as:
) ( X E Mean · (2.21)
The median, is defined as the value of X which is midway (in terms of
probability) between the smallest possible value and the largest possible
value. The median is the point, which divides the total area under the PDF
into two equal parts. In other words, the probability that X is less than the
median is1 2, and the probability that X is greater than the median is also
1 2. Thus, if P X a ( ) . ≤ ≥ 050 and P X a ( ) . ≥ ≥ 050 then 'a' is the
median of the distribution of X. In the continuous case, this can be expressed
as:
f x dx f x dx
a
a
( ) ( ) .
−∞
+∞
∫ ∫
· · 050 (2.22)
The mode, i s def i ned as t he val ue of X at w hi ch t he PDF of X r eaches i t s
hi ghest p oi nt . If a gr aph of t he PM F (PDF), or a l i st i ng of possi bl e val ues of X
al ong wi t h t hei r pr obabi l i t i es i s avai l abl e, det er mi nat i on of t he mode i s
qui t e si mpl e.
A cent r al t endency par amet er , whet her i t i s mode, medi an, mean, or any
ot her measur e, su mmar i ses onl y a cer t ai n aspect of a di st r i but i on. I t i s easy
t o f i nd t w o di st r i but i ons w hi ch have t he same mean but w hi ch ar e not at al l
si mi l ar i n any ot her r esp ect .
Measures of di spersi on
The mean i s a good i ndi cat i on of t he l ocat i on of a r an dom var i abl e, but no
si ngl e val ue need be exact l y l i ke t he mean. A devi at i on f r om t he mean, D,
expr esses t he measur e of er r or made by usi ng t he mean as a par t i cul ar
val ue:
2. Pr obabi l i t y Theor y 28
M x D − ·
Wher e, x, i s a possi bl e val ue of t he r andom var i abl e, X. The d evi at i on can
be t ak en f r om ot h er measur es of cent r al t end en cy such as t h e medi an or
mod e. It i s qui t e obvi ous t hat t he l ar ger such devi at i ons ar e f r o m a
measur e of cent r al t endency, t he mor e t he i ndi vi dual val ues di f f er f r om
each ot her , and t he mor e appar ent t he spr ead wi t hi n t he di st r i but i on
becomes. Consequent l y, i t i s necessar y t o f i nd a measur e t hat w i l l r ef l ect
t he spr ead, or var i abi l i t y, of i ndi vi dual val ues.
The expect at i on of t he devi at i on about t he mean as a measur e of
var i abi l i t y, E(X  M ), w i l l not w or k because t he expect ed d evi at i on f r o m t he
mean must be zer o f or obvi ous r easons. Th e sol ut i on i s t o f i nd t he squar e of
each d evi at i on f r o m t he mean, and t hen t o f i nd t h e expect at i on of t he
squar ed d evi at i on. Thi s char act er i st i c i s know n as a var i ance of t he
di st r i but i on, V, t hus:
V X E X Mean X Mean P x ( ) ( ) ( ) ( ) · − · − ×
∑
2 2
if X is discrete (2.23)
V X E X Mean X Mean f x dx ( ) ( ) ( ) ( ) · − · − ×
−∞
+∞
∫
2 2
if X is continuous
(2.24)
The posi t i ve squar e r oot of t he var i ance f or a di st r i but i on i s cal l ed t he
St andar d Devi at i on, SD.
) (X V SD · (2.25)
Pr obabi l i t y di st r i but i ons can be anal ysed i n gr eat er dept h b y i nt r oduci ng
ot her summar y measur es, known as moment s. Ver y si mpl y t h ese ar e
exp ect at i ons of di f f er ent pow er s of t he r andom var iabl e. M or e i nf or mat i on
about t hem can be f ound i n t ext s on pr obabi l i t y.
2. Pr obabi l i t y Theor y 29
Fi gur e 2.13 Pr obabi l i t y Syst em f or Cont i nuous Random Var i abl e
Vari abi l i t y
The st andar d devi at i on i s a measur e t hat show s how cl osel y t he val ues of
r andom var i abl es ar e concent r at ed ar ound t he mean. Somet i mes i t i s
di f f i cul t t o use onl y know l edge of t h e st andar d d evi at i on, t o d eci de wh et her
t he di sper si on i s consi der abl y l ar ge or smal l , because t hi s w i l l depend on
t he mean val u e. I n t hi s case t he par am et er know n as coef f i ci ent of
var i at i on, CV
X
, def i ned as
M
SD
CV
X
· (2.26)
Coef f i ci ent of var i at i on i s ver y usef ul because i t gi ves bet t er i nf or mat i on
r egar di ng t h e di sper si on. Th e concept t hus di scussed so f ar i s summar i sed
i n Fi gur e 2.13.
In concl usi on i t coul d b e sai d t hat t he pr obabi l i t y syst em i s w hol l y abst r act
and axi omat i c. Consequent l y, ever y f ul l y def i ned pr o babi l i t y pr obl em has a
uni que sol ut i on.
2.10. DISCRETE THEORETICALPROBABILITY
DISTRIBUTIONS
In pr obabi l i t y t heor y, t her e ar e sever al r ul es t hat def i ne t he f unct i onal
r el at i onshi p bet w een t he possi b l e val u es of r andom var i abl e X and t hei r
pr obabi l i t i es, P(X). As t hey ar e pur el y t heor et i cal , i . e. t hey do not exi st i n
r eal i t y, t hey ar e cal l ed t heor et i cal pr obabi l i t y di st r i but i ons. Inst ead of
anal ysi ng t he ways i n whi ch t hese r ul es have been der i ved, t he anal ysi s i n
t hi s chapt er con cent r at es on t hei r pr oper t i es. It i s necessar y t o emphasi se
t hat al l t heor et i cal di st r i but i ons r epr esent t he f ami l y of di st r i but i ons
def i ned by a co mmon r ul e t hr ough unspeci f i ed const ant s known as
par amet er s of di st r i but i on. The par t i cul ar member of t h e f ami l y i s d ef i ned
by f i xi ng numer i cal val ues f or t he par am et er s, whi ch def i ne t he di st r i but i on.
2. Pr obabi l i t y Theor y 30
The pr obabi l i t y di st r i but i ons most f r equent l y used i n r el i abi l i t y,
mai nt enan ce and t he l ogi st i c suppor t ar e exami ned i n t hi s chapt er .
Among t h e f ami l y of t h eor et i cal pr obabi l i t y di st r i but i ons t hat ar e r el at ed t o
di scr et e r andom var i abl es, t he Bi nomi al di st r i but i on and t he Poi sson
di st r i but i on ar e r el evan t t o t h e obj ect i ves set b y t hi s book. A br i ef
descr i pt i on of each now f ol l ows.
2.10.1 Bernuolli Trials
The si mpl e pr obabi l i t y di st r i but i on i s one wi t h onl y t wo event cl asses. For
exampl e, a car i s t est ed and one of t wo event s, pass or f ai l , must occur ,
each wi t h so me pr obabi l i t y. The t ype of exper i ment consi st i ng of ser i es of
i ndepen dent t r i al s, each of whi ch can event uat e i n onl y one of t wo
out comes ar e known as Ber nuol l i Tr i al s, and t he t w o event cl asses and t h ei r
associ at ed pr obabi l i t i es a Ber nuol l i Pr ocess. In gener al , one of t he t w o
event s i s cal l ed a “ success” and t h e ot her a “ f ai l ur e” or “ nonsuccess” .
These names ser ve onl y t o t el l t he event s apar t , and ar e not meant t o b ear
any connot at i on of “ goodness” of t he even t . The symbol p, st ands f or t he
pr obabi l i t y of a success, q f or t he pr obabi l i t y of f ai l ur e (p + q =1). If 5
i ndepen dent t r i al s ar e made (n = 5), t hen 2
5
= 32 di f f er ent sequences of
possi bl e out comes woul d be obser ved.
The pr obabi l i t y of gi ven sequen ces d epends upon p and q, t h e pr obabi l i t y of
t he t wo event s. For t unat el y, si nce t r i al s ar e i ndepen dent , i t i s possi bl e t o
comput e t h e pr obabi l i t y of any sequence.
If al l possi bl e sequences and t hei r pr obabi l i t i es, ar e w r i t t en dow n t he
f ol l owi ng f act emer ges: The pr obabi l i t y of any gi ven sequences of n
i ndependent Ber nuol l i Tr i al s depends onl y on t he number of successes and
p. Thi s i s r egar dl ess of t he or der i n w hi ch successes and f ai l ur e occur i n
sequence, t he pr obabi l i t y i s
p q
r n r −
where r is the number of successes, and n r − is the number of failures.
Suppose that in a sequence of 10 trials, exactly 4 success occurs. Then the
probability of that particular sequence is p q
4 6
. If
3
2
· p , then the
probability can worked out from:
2
3
1
3
4 6
¸
¸
_
,
¸
¸
_
,
2. Pr obabi l i t y Theor y 31
The sam e pr ocedur e w oul d be f ol l ow ed f or any r successes out of n t r i al s
f or any p. Gen er al i si ng t hi s i dea f or any r , n, and p, w e have t he f ol l ow i ng
pr i nci pl e:
In sampling from the Bernuolli Process with the probability of a success
equal to p, the probability of observing exactly r successes in n independent
trials is:
P r successes n p
n
r
p q
n
r n r
p q
r n r r n r
(  , )
!
!( )!
·
¸
¸
_
,
·
−
− −
(2.27)
2.10.2 The Binomial Distribution
The theoretical probability distribution, which pairs the number of successes
in n trials with its probability, is called the binominal distribution.
Thi s pr obabi l i t y di st r i but i on i s r el at ed t o exp er i ment s, w hi ch consi st of a
ser i es of i ndep endent t r i al s, each of whi ch can r esul t i n onl y on e of t wo
out comes: success and or f ai l u r e. These names ar e used onl y t o t el l t he
event s apar t . By conven t i on t h e symbol p st ands f or t he pr obabi l i t y of a
success, q f or t he pr obabi l i t y of f ai l ur e ( ) p q + · 1 .
The number of successes, x i n n t r i al s i s a di scr et e r andom var i abl e whi ch
can t ake on onl y t he whol e val ues f r o m 0 t hr ough n. The PM F of t he
Bi nomi al di st r i but i on i s gi ven by:
PMF x P X x
n
x
p q x n
x n x
( ) ( ) , · · ·
¸
¸
_
,
< <
−
0 (2.28)
wher e:
n
x
p q
n
x n x
p q
x n x x n x
¸
¸
_
,
·
−
− −
!
!( )!
(2.29)
The bi nomi al di st r i but i on expr essed i n cu mul at i ve f o r m, r epr esen t i ng t he
pr obabi l i t y t hat X f al l s at or bel ow a cer t ai n val ue ' a' i s def i ned by t he
f ol l ow i ng equat i on:
2. Pr obabi l i t y Theor y 32
P X a P X x
n
i
p q
i
i n i
i
a
i o
a
( ) ( ) ≤ · · ·
¸
¸
_
,
−
· ·
∑ ∑
0
(2.30)
As an i l l ust r at i on of t he bi nomi al di st r i but i on, t he PM F and CDF ar e sh own
i n Fi gur e 2.14 wi t h par amet er s n = 10 and p = 0. 3.
Figure 2.14 PMF and CDF For Binomial Distribution, n = 10, p = 0.3
E X np ( ) · (2.31)
Si mi l ar l y, because of t he i ndep endence of t r i al s, t he var i ance of t he
bi nomi al di st r i but i on i s t he sum of t he var i ances of t he i ndi vi dual t r i al s, or
p p ( ) 1− summed n t i mes:
V X np p npq ( ) ( ) · − · 1 (2.32)
Conseq uent l y, t he st andar d devi at i on i s equal t o:
2. Pr obabi l i t y Theor y 33
Sd X npq ( ) · (2.33)
Al t hough t he mat hemat i cal r ul e f or t he bi nomi al di st r i but i on i s t he same
r egar dl ess of t he par t i cul ar val ues w hi ch par amet er s n and p t ake, t he
shape of t he pr obabi l i t y mass f unct i on and t he cu mul at i ve di st r i but i on
f unct i on w i l l depend upon t hem. The PM F of t he bi nomi al di st r i but i on i s
symmet r i c i f p = 0.5, posi t i vel y skew ed i f p < 0.5, and negat i vel y skew ed i f p
> 0. 5.
2.10.3 The Poisson Distribution
The t heor et i cal pr obabi l i t y di st r i but i on whi ch pai r s t he number of
occur r ences of an event i n a gi ven t i me per i od w i t h i t s pr obabi l i t y i s cal l ed
t he Poi sson di st r i but i on. Ther e ar e exp er i ment s wher e i t i s not possi bl e t o
obser ve a f i ni t e sequence of t r i al s. I nst ead, obser vat i ons t ake pl ace o ver a
cont i nuum, such as t i me. For exampl e, i f t he number of car s ar ri vi ng at a
speci f i c j unct i on i n a gi ven per i od of t i me i s obser ved, say f or on e mi nut e, i t
i s di f f i cul t t o t hi nk of t hi s si t uat i on i n t er ms of f i ni t e t r i al s. I f t he number of
bi nomi al t r i al s n, i s made l ar ger and l ar ger and p smal l er and smal l er i n such
a w ay t hat np r emai ns const ant , t hen t h e pr obabi l i t y di st r i but i on of t he
number of occur r ences of t he r andom var i abl e appr oaches t he Poi sson
di st r i but i on.
The pr obabi l i t y mass f unct i on i n t he case of t he Po i sson di st ri but i on f or
r andom var i abl e X can be expr essed as f ol l o w s:
P X x
e
x
x
(  )
!
· ·
−
λ
λ
λ
where x = 0, 1, 2, . (2.34)
λ i s t he i nt ensi t y of t he pr ocess and r epr esent s t h e expect ed number of
occur r ences i n a t i me per i od of l engt h t . Fi gur e 2.15 show s t he PM F of t he
Poi sson di st r i but i on wi t h λ · 5
2. Pr obabi l i t y Theor y 34
Figure 2.15 PMF of the Poisson Distribution with λ · 5
The Cumul at i ve Di st r i but i on Funct i on f or t he Poi sson di st r i but i on
F x P X x
e
i
i x
i o
x
( ) ( )
!
· ≤ ·
·
∑
λ
(2.35)
The CDF of t he Poi sson di st r i but i on w i t h λ · 5 i s pr esent ed i n Fi gur e 2.16.
Expect ed val ue of t he di st r i but i on i s gi ven by
E X xP X x x
e
x
x
x
x
( ) ( )
!
· · ·
·
−
·
∑ ∑
0 0
λ
λ
Appl yi ng some si mpl e mat h emat i cal t r ansf or mat i ons i t can be pr oved t hat :
E X ( ) · λ (2.36)
whi ch means t hat t he exp ect ed number of occur r ences i n a per i od of t i me t
i s equal t o np, whi ch i s equal t o λ .
The var i ance of t he Poi sson di st r i but i on i s equal t o t he mean:
V X ( ) · λ (2.37)
Thus, t he Poi sson di st r i but i on i s a si ngl e par amet er di st r i but i on because i t i s
compl et el y d ef ined by t he par amet er λ . In gener al , t he Poi sson
2. Pr obabi l i t y Theor y 35
di st r i but i on i s posi t i vel y skew ed, al t hough i t i s n ear l y symmet r i cal as
λ beco mes l ar ger .
2. Pr obabi l i t y Theor y 36
Fi gur e 2. 16 CDF of t he Poi sson Di st r i but i on λ · 5
The Poi sson di st r i but i on can b e der i ved as a l i mi t i ng f or m of t h e bi nomi al i f
t he f ol l owi ng t hr ee assumpt i ons wer e si mul t aneousl y sat i sf i ed:
1. n becomes large (that is, n → ∞).
2. p becomes small (that is, p →0).
3. np remains constant.
Under t hese condi t i ons, t he bi nomi al di st r i but i on wi t h t he par amet er s n
and p, can be appr oxi mat ed t o t he Poi sson di st r i but i on wi t h
par amet er λ · np . Thi s means t hat t he Poi sson di st r i but i on pr ovi d es a
good appr oxi mat i on t o t he bi nomi al di st r i but i on i f p i s ver y smal l and n i s
l ar ge. Si nce p and q can be i nt er changed b y si mp l y i nt er changi ng t he
def i ni t i ons of success and f ai l u r e, t he Poi sson di st r i but i on i s al so a good
appr oxi mat i on when p i s cl ose t o one and n i s l ar ge.
As an exampl e of t h e use of t he Poi sso n di st r i but i on as an appr oxi mat i on t o
t he bi nomi al di st r i but i on, t he case i n w hi ch n = 10 and p = 0.10 w i l l be
consi der ed. The Poi sson par amet er f or t h e ap pr oxi mat i on i s t hen
λ · · × · np 10 010 1 . . The bi nomi al di st r i but i on and t he Poi sson
appr oxi mat i on ar e shown i n Tabl e 2.2.
The t w o di st r ibut i ons agr ee r easonabl y w el l . If mor e pr eci si on i s desi r ed, a
possi bl e r ul e of t humb i s t hat t he Poi sson i s a good appr oxi mat i on t o t he
bi nomi al i f n p / > 500 (t hi s shoul d gi ve accur acy t o at l east t w o deci mal
pl aces).
2. Pr obabi l i t y Theor y 37
Tabl e 2.2 Poi sson Di st r i but i on as an Appr oxi mat ion t o t he Bi nomi al
Di st r i but i on
Binomial
P X x n p (  , . ) · · · 10 01
Poisson
P X x (  ) · · λ 1
0 0.598737 0.606531
1 0.315125 0.303265
2 0.074635 0.075816
3 0.010475 0.012636
4 0.000965 0.001580
5 0.000061 0.000158
2.11. CONTINUOUS THEORETICAL PROBABILITY
DISTRIBUTIONS
It i s necessar y t o emphasi se t hat al l t heor et i cal di st r i but i ons r epr esent t he
f ami l y of di st r i but i ons d ef i ned b y a common r u l e t hr ou gh unspeci f i ed
const ant s known as par amet er s of di st r i but i on. The par t i cul ar member of
t he f ami l y i s def i ned b y f i xi ng numer i cal val ues f or t he par amet er s, w hi ch
def i ne t he di st r i but i on. The pr obabi l i t y di st r i but i ons most f r equent l y used
i n r el i abi l i t y, mai nt ai nabi l i t y and suppor t abi l i t y en gi neer i ng ar e exami ned i n
t hi s chapt er . Each of t h e abo ve ment i oned r ul es def i ne a f ami l y of
di st r i but i on f unct i ons. Each member of t he f ami l y i s def i ned wi t h a f ew
par amet er s, w hi ch i n t hei r ow n w ay cont r ol t he di st r i but i on. Par amet er s of
a di st r i but i on can b e cl assi f i ed i n t he f ol l owi ng t hr ee cat egor i es (not e t hat
not al l di st r i but i ons wil l have al l t he t hr ee par amet er s, many di st r i but i ons
may have ei t her one or t wo par amet er s):
1. Scale parameter, which controls the range of the distribution on the
horizontal scale.
2. Shape parameter, which controls the shape of the distribution curves.
3. Source parameter or Location parameter, which defines the origin or the
minimum value which random variable, can have. Location parameter
also refers to the point on horizontal axis where the distribution is
located.
Thus, i ndi vi dual member s of a speci f i c f ami l y of t he pr obabi l i t y di st r i but i on
ar e def i ned b y f i xi ng numer i cal val ues f or t he above par amet er s.
2. Pr obabi l i t y Theor y 38
2.11.1 Exponential Distribution
Exponent i al di st r i but i on i s f ul l y def i ned b y a si ngl e one par amet er t hat
gover ns t he scal e of t he di st r i but i on. The pr obabi l i t y d ensi t y f unct i on of t he
exponent i al di st r i but i on i s gi ven by:
, ) 0 , exp ) ( > − · x x x f λ λ (2.38)
In Fi gur e 2.17 sever al gr aphs ar e shown of exponen t i al densi t y f unct i ons
w i t h di f f er ent val ues of λ. Not i ce t hat t he expon ent i al di st r i but i on i s
posi t i vel y skew ed, wi t h t he mode occur r i ng at t h e smal l est possi bl e val ue,
zer o.
Fi gur e 2.17. Pr obabi l i t y densi t y f unct i on of exponent i al di st r i but i on f or
di f f er en t val ues of λ
The cu mul at i ve di st r i but i on of exponen t i al di st r i but i on i s gi ven by:
, ) , ) x x X P x F λ − − · < · exp 1 ) ( ) ( (2.39)
It can be show n t hat t he mean and var i ance of t he exponent i al di st r i but i on
ar e:
λ / 1 ) ( · X E (2.40)
2
) / 1 ( ) ( λ · X V (2.41)
2. Pr obabi l i t y Theor y 39
The standard deviation in the case of the exponential distribution rule has a
numerical value identical to the mean and the scale parameter,
λ / 1 ) ( ) ( · · X E X SD .
11.1.1 Memoryless Property of Exponential Distribution
One of t h e uni que pr op er t y of exponen t i al di st r i but i on i s t hat i t i s t he onl y
cont i nuous di st r i but i on t hat has memor y l ess pr oper t y. Suppose t hat t he
r andom var i abl e X measur es t he dur at i on of t i me u nt i l t he occur r ence of
f ai l ur e of an i t em and t hat i t i s known t hat X has an exponen t i al di st r i but i on
wi t h par amet er λ. Suppose t he pr esent age of t he i t em i s t , t hat i s X > t .
Assume t hat w e ar e i nt er est ed i n f i ndi ng t he pr obabi l i t y t hat t hi s i t em wi l l
not f ai l f or anot her s uni t s of t i me. Thi s can be expr essed usi ng t he
condi t i onal pr obabi l i t y as:
} { t x t s X P > + >
Usi ng condi t i onal pr obabi l i t y of even t s, t he above pr obabi l i t y can be
w r i t t en as:
} {
} {
} {
} {
} {
t X P
t s X P
t X P
t X t s X P
s X t s X P
>
+ >
·
>
> ∩ + >
· > + > (2.42)
How ever w e know t hat f or exponent i al di st r i but i on
)) ( exp( ] [ t s t s X P + − · + > λ and ) exp( ] [ t t X P λ − · >
Subst i t ut i ng t hese expr essi ons i n equat i on (2.42), we get
) exp( ] [ ] [ s s X P t X t s X P λ − · > · > + >
That i s, t he condi t i onal pr obabi l i t y d epends onl y on t h e r emai ni ng d ur at i on
and i s i ndependent of t he cur r ent age of t he i t em. Thi s pr oper t y i s expl oi t ed
t o a gr eat ext end i n r el i abi l i t y t heor y.
2.11.2 Normal Distribution (Gaussian Distribution)
Thi s i s t he most f r equ ent l y used and most ext ensi vel y cover ed t heor et i cal
di st r i but i on i n t he l i t er at ur e. Th e Nor mal Di st r i but i o n i s cont i nuous f or al l
val ues of X bet w een − ∞ and + ∞. It has a char act er i st i c symmet r i cal
2. Pr obabi l i t y Theor y 40
shape, w hi ch means t hat t he mean, t he medi an and t he mode have t he
same numer i cal val ue. Th e mat hemat i cal expr essi on f or i t s pr obabi l i t y
densi t y f unct i on i s as f ol l ow s:
,
_
¸
¸
,
_
¸
¸ −
− ·
2
2
1
exp
2
1
) (
σ
µ
π σ
x
x f (2.43)
Wh er e µ i s a l ocat i on par amet er (as i t l ocat es t he di st r i but i on on t he
hor i zont al axi s) and σ i s a scal e par amet er (as i t cont r ol s t he r ange of t he
di st r i but i on). µ and σ al so r epr esent s t he mean and t he st andar d devi at i on
of t hi s di st r i but i on.
The i nf l uence of t he par amet er µ on t he l ocat i on of t he di st r i but i on on t he
hor i zont al axi s i s shown i n Fi gur e 2.18, wh er e t he val ues f or par am et er σ
ar e const ant .
As t he devi at i on of x f r om t he l ocat i on par amet er µ i s ent er ed as a squar ed
quant i t y, t wo di f f er ent x val ues, show i ng t h e same absol ut e devi at i on f r om
µ, wi l l have t he same pr obabi l i t y densi t y accor di ng t o t hi s r ul e. Thi s di ct at es
t he symmet r y of t he nor mal di st r i but i on. Par amet er µ can be any f i ni t e
number , whi l e σ can be any posi t i ve f i ni t e number .
The cumul at i ve di st r i but i on f unct i on f or t he nor mal di st r i but i on i s:
F a P X a f x dx
a
( ) ( ) ( ) · ≤ ·
−∞
∫
wher e f (x) i s t he nor mal densi t y f unct i on. Taki ng i nt o accoun t Eq. (2.43)
t hi s becomes:
dx
a
a F
a
∫
,
_
¸
¸
,
_
¸
¸ −
− ·
∞ −
2
2
1
exp
2
1
) (
σ
µ
π σ
(2.44)
2. Pr obabi l i t y Theor y 41
Fi gur e 2.18 Pr obabi l i t y densi t y of nor mal di st r i but i on f or di f f er ent σ val ues
In Fi gur e 2.19 sever al cu mul at i ve di st r i but i on f unct i ons ar e gi ven of t he
Nor mal Di st r i but i on, cor r espondi ng t o di f f er ent val u es of µ and σ .
As t he i nt egr al i n Eq. (2.44) cannot be eval uat ed i n a cl osed f or m,
st at i st i ci ans have const r uct ed t h e t abl e of pr obabi l i t i es, whi ch compl i es
wi t h t he nor mal r ul e f or t he st andar di sed r andom var i abl e, Z. Thi s i s a
t heor et i cal r andom var i abl e wi t h par amet er s µ = 0 and σ = 1. The
r el at i onshi p bet w een st andar di sed r andom var i abl e Z and r andom var i abl e
X i s est abl i shed by t he f ol l owi ng expr essi on:
Fi gur e 2.19 Cumul at i ve di st r i but i on of nor mal di st r i but i on f or di f f er ent
val ues of µ and σ.
σ
µ −
·
x
z (2.45)
2. Pr obabi l i t y Theor y 42
M aki ng use of t he abo ve expr essi on t he eq uat i on (2.43) becomes
si mpl er :
2
2
1
2
1
) (
z
e z f
−
·
π σ
(2.46)
The st andar di sed f or m of t he di st r i but i on makes i t possi bl e t o use onl y one
t abl e f or t he d et er mi nat i on of PDF f or any nor mal di st r i but i on, r egar dl ess of
i t s par t i cul ar par amet er s (see Tabl e i n appendi x).
The r el at i onshi p bet w een f (x) and f (z) i s :
σ
) (
) (
z f
x f · (2.47)
By substituting
σ
µ − x
with z Eq. (2.44) becomes:
,
_
¸
¸ −
Φ ·
∫
,
_
¸
¸
− ·
∞ −
σ
µ
π σ
x
dz z a F
z
2
2
1
exp
2
1
) ( (2.48)
w her e Φ i s t he st andar d nor mal di st r i but i on Funct i on def i ned by
Φ( ) exp z z dx
x
· −
¸
¸
_
,
−∞
∫
1
2
1
2
2
π
(2.49)
The cor r espondi ng st andar d nor mal pr obabi l i t y densi t y f unct i on i s:
f z
z
( ) exp · −
¸
¸
_
,
1
2 2
2
π
(2.50)
M ost t abl es of t he nor mal di st r i but i on gi ve t he cumul at i ve pr obabi l i t i es f or
var i ous st andar di sed val u es. That i s, f or a gi ven z val ue t he t abl e pr o vi des
t he cu mul at i ve pr obabi l i t y up t o, and i ncl udi ng, t hat st andar di sed val u e i n a
nor mal di st r i but i on. In M i cr osof t EXCEL
©
, t he cumul at i ve di st r i but i on
2. Pr obabi l i t y Theor y 43
f unct i on and densi t y f unct i on of nor mal di st r i but i on wi t h mean µ and
st andar d devi at i on σ can be f ound usi ng t he f ol l ow i ng f unct i on.
F(x) = NORM DI ST (x, µ, σ, TRUE), and f (x) = NORM DI ST (x, µ, σ, FALSE)
The expect at i on of a r andom var i abl e, i s equal t o t he l ocat i on par amet er µ
t hus:
µ · ) ( X E (2.51)
Wh er eas t he var i ance i s
2
) ( σ · X V (2.52)
Si nce nor mal di st r i but i on i s a symmet r i cal about i t s mean, t he ar ea
bet w een µ  kσ, µ + kσ (k i s any r eal number ) t akes a uni que val ue, w hi ch i s
shown i n Fi gur e 2.20.
Fi gur e 2. 20 The ar eas under a nor mal di st r i but i on bet w een
µ  kσ and µ + kσ
11.2.1 Central Limit Theorem
Suppose X
1
, X
2
, … X
n
ar e mut ual l y i ndep end ent obser vat i ons on a r andom
var i abl e X havi ng a wel l def i ned mean µ
x
and st andar d devi at i on σ
x
. Let
n
X
Z
x
x
n
/ σ
µ −
·
−
(2.53)
Wher e,
2. Pr obabi l i t y Theor y 44
∑ ·
·
−
n
i
i
X
n
X
1
1
(2.54)
and ) (z F
n
z
be the cumulative distribution function of the random variable
Z
n
. Then for all z,  ∞ < z < ∞,
) ( ) ( lim z F z F
Z Z
n
n
·
∞ →
(2.55)
wher e F
Z
(z) i s t he cumul at i ve di st r i but i on of st andar d nor mal di st r i but i on
N(0,1). The X val ues have t o be f r om t h e same di st r i but i on but t he
r emar kabl e f eat ur e i s t hat t hi s di st r i but i on does no t have t o be nor mal , i t
can be uni f or m, exponent i al , bet a, gamma, Wei bul l or even an u nknown
one.
2.11.3 Lognormal Distribution
The l ognor mal pr obabi l i t y di st r i but i on, can i n some r espect s, be consi der ed
as a speci al case of t he nor mal di st r i but i on because o f t he der i vat i on of i t s
pr obabi l i t y f unct i on. If a r andom var i abl e Y X · ln i s nor mal l y di st r i but ed
t hen, t h e r andom var i abl e X f ol l ows t he l ognor mal d i st r i but i on. Thus, t he
pr obabi l i t y densi t y f unct i on f or a r andom var i abl e X i s def i ned as:
0
ln
2
1
exp
2
1
) (
2
≥
,
_
¸
¸
,
_
¸
¸ −
− ·
l
l
l
X
x
x
x f
σ
µ
π σ
(2.56)
The par amet er
l
µ i s cal l ed t he scal e par amet er (see Fi gur e 2.21) and
par amet er
l
σ i s cal l ed t he shape par amet er . Th e r el at i onshi p bet w een
par amet er s µ (l ocat i on par amet er of t he nor mal di st r i but i on) and
l
µ i s
def i ned:
,
_
¸
¸
+ ·
2
2
1
exp
l
l
σ µ µ (2.57)
The cumul at i ve di st r i but i on f unct i on f or t he l ognor mal di st r i but i on i s
def i ned wi t h t he f ol l owi ng expr essi on:
2. Pr obabi l i t y Theor y 45
Fi gur e 2.21 Pr obabi l i t y densi t y of l ognor mal di st r i but i on
dx
x
x
x X P x F
x
l
l
l
X
∫
,
_
¸
¸
,
_
¸
¸ −
− · ≤ ·
0
2
ln
2
1
exp
2
1
) ( ) (
σ
µ
π σ
(2.58)
As t he i nt egr al cannot b e eval uat ed i n cl ose f or m, t he same pr ocedur e i s
appl i ed as i n t h e case of nor mal di st r i but i on. Then , maki ng use of t he
st andar di sed r andom var i abl e Equat i on (2.61) t r ansf or ms i nt o:
,
_
¸
¸ −
Φ · ≤ ·
l
l
X
x
x X P x F
σ
µ ln
) ( ) ( (2.59)
The measur es of cent r al t endency i n t h e case of l ognor mal di st r i but i ons ar e
def i ned by t he:
(a) Locat i on par amet er (M ean)
,
_
¸
¸
+ · ·
2
2
1
exp ) (
l
l
X E M σ µ (2.60)
(b) Devi at i on par amet er (t he var i ance)
2. Pr obabi l i t y Theor y 46
, ) ,  ) 1 exp( 2 exp ) (
2
2
− + ·
l l
l
X V σ σ µ (2.61)
2.11.4 Weibull Distribution
This distribution originated from the experimentally observed variations in
the yield strength of Bofors steel, the size distribution of fly ash, fibre
strength of Indian cotton, and the fatigue life of a St37 steel by the Swedish
engineer W.Weibull. As the Weibull distribution has no characteristic shape,
such as the normal distribution, it has a very important role in the statistical
analysis of experimental data. The shape of this distribution is governed by
its parameter.
The rule for the probability density function of the Weibull distribution is:
1
1
]
1
¸
,
_
¸
¸ −
−
,
_
¸
¸ −
·
− β β
η
γ
η
γ
η
β x x
x f exp ) (
1
(2.65)
where η, β, γ > 0. As the location parameter ν is often set equal to zero, in
such cases:
1
1
]
1
¸
,
_
¸
¸
−
,
_
¸
¸
·
− β β
η η η
β x x
x f exp ) (
1
(2.66)
By altering the shape parameter β, the Weibull distribution takes different
shapes. For example, when β = 3.4 the Weibull approximates to the normal
distribution; when β =1, it is identical to the exponential distribution. Figure
2.22 shows the Weibull probability density function for selected parameter
values.
The cumul at i ve di st r i but i on f unct i ons f or t he Wei bul l di st r i but i on i s:
2. Pr obabi l i t y Theor y 47
1
1
]
1
¸
,
_
¸
¸ −
− − ·
B
x
x F
η
γ
exp 1 ) ( (2.67)
Fi gur e 2.22. Pr obabi l i t y densi t y of Wei bul l di st r i but i on wi t h β = 2.0,
γ = 0, η = 0.5, 1, 2
For γ = 0, the cumulative distribution is given by
1
1
]
1
¸
,
_
¸
¸
− − ·
β
η
x
x F exp 1 ) ( (2.68)
The expect ed val ue of t he Wei bul l di st r i but i on i s gi ven by:
,
_
¸
¸
+ Γ × + · 1
1
) (
β
η γ X E (2.69)
wher e Γ i s t he gamma f unct i on, def i ned as
2. Pr obabi l i t y Theor y 48
dx x e n
n x 1
0
) (
−
∞
−
×
∫
· Γ
When n is integer then )! 1 ( ) ( − · Γ n n . For other values, one has to solve
the above integral to the value. Values for this can be found in Gamma
function table given in the appendix. In Microsoft EXCEL, Gamma function,
) (x Γ can be found using the function, EXP[GAMMALN(x)].
The var i ance of t he Wei bul l di st ri but i on is gi ven by:
1
]
1
¸
,
_
¸
¸
+ Γ −
,
_
¸
¸
+ Γ ·
β β
η
1
1
2
1 ) ( ) (
2 2
X V (2.70)
3. Rel i abi l i t y M easur es 49
Chapter 3
Reliability Measures
I have seen t he f ut ur e; and i t w or ks
Li ncol n St ef f ens
In t hi s chapt er we di scuss var i ous measur es by whi ch har dwar e and
sof t war e r el i abi l i t y char act er i st i cs can b e numer i cal l y def i ned and
descr i b ed. M anuf act ur er s and cust omer s use r el i abi l i t y measur e t o quant i f y
t he ef f ect i ven ess of t h e syst em. Use of any par t i cul ar r el i abi l i t y measur e
dep ends on what i s expect ed of t he syst em and what we ar e t r yi ng
measur e. Sever al l i f e cycl e d eci si on ar e mad e usi n g r el i abi l i t y measur e as
one of t he i mpor t ant desi gn par amet er . The r el i abi l i t y char act er i st i cs or
measur es used t o sp eci f y r el i abi l i t y must r ef l ect t he op er at i onal
r equi r ement s of t h e i t em. Requi r ement s must b e t ai l or ed t o i ndi vi dual i t em
consi der i ng oper at i onal envi r onmen t and mi ssi on cr i t i cal i t y. In br oader
sense, t he r el i abi l i t y met r i cs can be cl assi f i ed (Fi gur e 3.1) as: 1. Basi c
Rel i abi l i t y M easur es, 2. M i ssi on Rel i abi l i t y M easur es, 3. Oper at i onal
Rel i abi l i t y M easur es, and 4. Cont r act ual Rel i abi l i t y M easur es.
Basi c Rel i abi l i t y M easur es ar e used t o pr edi ct t he syst em' s abi l i t y t o op er at e
wi t hout mai nt enan ce and l ogi st i c suppor t . Rel i abi l i t y measur es l i ke
r el i abi l i t y f unct i on and f ai l ur e f unct i on f all under t hi s cat egor y.
M i ssi on Rel i abi l i t y M easur es ar e used t o pr edi ct t h e syst em' s abi li t y t o
compl et e mi ssi on. These measur es consi der onl y t hose f ai l ur es t hat cause
mi ssi on f ai l ur e. Rel i abi l i t y measur es such as mi ssi on r el i abi l i t y,
mai nt enance f r ee oper at i ng per i od (M FOP), f ai l ur e f r ee oper at i ng p er i od
(FFOP), and hazar d f unct i on f al l under t hi s cat egor y.
Oper at i onal Rel i abi l i t y M easur es ar e used t o pr edi ct t h e per f or mance of t he
syst em wh en oper at ed i n a pl anned envi r onment i ncl udi ng t he combi ned
ef f ect of desi gn, qual i t y, envi r onment , mai nt enan ce, suppor t pol i cy, et c.
3. Rel i abi l i t y M easur es 50
M easur es such as M ean Ti me Bet ween M ai nt enance (M TBM ), M ean Ti me
Bet w een Over haul (M TBO), M ai nt enance Fr ee Op er at i ng Per i od (M FOP),
M ean Ti me Bet ween Cr i t i cal Fai l ur e (M TBCF) and M ean Ti me Bet ween
Unschedul ed Removal (M TBUR) f al l under t hi s cat egor y.
Cont r act ual Rel i abi l i t y M easur e i s used t o def i ne, measur e and eval uat e t he
manuf act ur er ' s pr ogr am. Cont r act ual r el i abi l i t y i s cal cul at ed by consi der i ng
desi gn and manuf act ur i ng char act er i st i cs. Basi cal l y i t i s t he i nher ent
r el i abi l i t y char act er i st i c. M easur es such as M ean Ti me To Fai l ur e (MTTF),
M ean Ti me Bet w een Fai l ur e (M TBF) and Fai l ur e r at e f al l under t hi s
cat egor y.
Fi gur e 3.1 Cl assi f i cat i ons of Rel i abi l i t y M easur es
Though w e cl assi f y t he r el i abi l i t y measur es i nt o f our cat egor i es as
men t i oned above, on e may r equi r e mor e t han one r el i abi l i t y met r i c i n most
of t h e cases f or speci f yi ng r el i abi l i t y r equi r emen t s. Sel ect i on of sp eci f i c
measur e t o quant i f y t he r el i abi l i t y r equi r emen t s shoul d i ncl ude mi ssi on and
l ogi st i c r el i abi l i t y al ong wi t h mai nt enance and suppor t measur es. Cur r ent l y,
many manuf act ur er s speci f y r el i abi l i t y by usi ng mean t i me bet w een f ai l ur e
(M TBF) and f ai l ur e r at e. However , M TBF and f ai l ur e r at es have sever al
dr awbacks. Recent pr oj ect s such as Fut ur e Of f ensi ve Ai r Syst ems (FOAS)
dr i ve mai nt enance f r ee oper at i ng per i ods (M FOP) as t he pr ef er r ed
r el i abi li t y r equi r ement .
In t he next Sect i on, we d ef i ne var i ous r el i abi l i t y measur es and how t o
eval uat e t hem i n pr act i cal pr obl ems. Al l t he measur es ar e d ef i ned based on
t he assump t i on t hat t he t i met of ai l ur e (TTF) di st r i but i on of t he syst em i s
know n. Pr ocedur es f or f i ndi ng t h e t i met of ai l ur e di st r i but i on by anal ysi ng
t he f ai l ur e dat a t hat ar e di scussed i n Chapt er 12.
Reliability Measures
Basic Reliability
Mission Reliability
Operational Reliability Contractual Reliability
3. Rel i abi l i t y M easur es 51
3.12. FAILURE FUNCTION
Fai l ur e f unct i on i s a basi c (l ogi st i c) r el i abi l i t y measur e and i s def i ned as t he
pr obabi l i t y t hat an i t em wi l l f ai l bef or e or at t h e moment of oper at i ng t i me
t . Her e t i me t i s used i n a gener i c sense and i t can have uni t s such as mi l es,
number of l andi ngs, f l yi ng hour s, number of cycl es, et c., dependi ng on t he
oper at i onal pr of il e and t he ut i l i sat i on of t he syst em. That i s, Fai l ur e f unct i on
i s equal t o t he pr obabi l i t y t hat t he t i met of ai l ur e r andom var i abl e w i l l be
l ess t han or equal a par t i cul ar val ue t (i n t hi s case oper at i ng t i me, see Fi gur e
3.2a). The f ai l ur e f unct i on i s usual l y r epr esent ed as F(t ).
F(t ) = P (f ai l ur e wi l l occur bef or e or at t i me t ) = P (TTF ≤ t )
= du u f
t
∫
0
) ( (3.1)
Fi gur e 3.2a. Fai l ur e f unct i on of a hypot het i cal di st r i but i on
Wh er e ) (t f i s t he pr obabi l i t y d ensi t y f unct i on of t he t i met of ai l ur e
r andom var i abl e TTF. Exponent i al , Wei bul l , nor mal , l ognor mal , Gamma and
Gumbel ar e f ew popul ar t heor et i cal di st r i but i ons t hat ar e used t o r epr esent
f ai l ur e f unct i on. Equat i on (3.1) i s der i ved by assumi ng t hat no mai nt enance
i s per f or med t o t h e syst em, and gi ves t he pr obabi l i t y of f ai l ur e f r ee
oper at i on wi t hout mai nt enance up t o t i me t . However , most of t he
compl ex syst ems wi l l r equi r e mai nt enance at f r equent i nt er val s. In such
cases, equat i on (3.1) has t o be modi f i ed, t o i ncor por at e t he b ehavi our of
t he syst em und er mai nt enance. Fai l ur e f unct i ons of f ew popul ar
t heor et i cal di st r i but i ons ar e li st ed i n Tabl e 3.1.
It shoul d be not ed t hat i n case of nor mal di st r i but i on t he f ai l ur e f unct i on
exi st s bet w een ∞ and +∞, so may have si gni f i cant val ue at t ≤ 0. Si nce
negat i ve t i me i s meani ngl ess i n r el i abi l i t y, gr eat car e shoul d be t aken i n
usi ng nor mal di st r i but i on f or t he f ai l ur e f unct i on. For µ >> 3σ, pr obabi l i t y
val ues f or t ≤ 0 can be consi der ed negl i gi bl e.
Ti me
F(t )
f t ( )
F
a
i
l
u
r
e
d
e
n
s
i
t
y
3. Rel i abi l i t y M easur es 52
Table 3.1 Failure function, F(t), of few theoretical distributions
Distribution Failure Function, F(t)
Exponential
1 0 0 − − > > exp( ) , λ λ t t
Normal
∫
,
_
¸
¸ −
Φ
,
_
¸
¸ −
− t
x
t
or dx e
0
]
2
1
[
2
2
1
σ
µ
π σ
σ
µ
or NORM DIST(t , µ, σ, TRUE) i n EXCEL
®
Lognormal
∫
,
_
¸
¸ −
Φ
,
_
¸
¸
,
_
¸
¸ −
−
t
l
l
x
l
t
or dx e
x
l
l
0
) ln(
2
1
) ln(
2
1
2
σ
µ
π σ
σ
µ
or NORM DIST(l n(t ), µ, σ, TRUE) i n EXCEL
®
Weibull 1 0 − −
−
> ≥ exp( ( ) ) , , ,
t
t
γ
η
η β γ γ
β
Gamma
1
1
0
Γ( ) α
β
α α β
x e dx
x
t
− −
∫
Not e t hat t h e f ai l ur e f unct i on of nor mal di st r i but i on i s def i ned b et ween 0
and t , si nce t i s gr eat er t han 0 f or r el i abi l i t y pur poses (agai nst t he usual l i mi t
∞) Appl i cat i ons of f ai l ur e f unct i on ar e l i st ed bel ow (Fi gur e 3.2b). Fai l ur e
f unct i ons of var i ous t heor et i cal di st r i but i ons f or di f f er ent par amet er val ues
ar e shown i n Fi gur es 3.3a3.3c.
Charact eri st i cs of f ai l ure f unct i on
1. Failure function is an increasing function. That is, for t
1
<t
2
, F (t
1
) ≤ F (t
2
).
2. For modelling purposes it is assumed that the failure function value at
time t = 0, F(0) = 0. However, this assumption may not be valid always.
3. Rel i abi l i t y M easur es 53
For example, systems can be dead on arrival. The value of failure
function increases as the time increases and for t = ∞, F(∞) = 1.
Appl i cat i ons of f ai l ure f unct i on
1. F(t) is the probability that an individual item will fail by time t.
2. F(t) is the fraction of items that fail by time t.
3. 1  F(t) is the probability that an individual item will survive up to time t.
Figure 3.2b. Properties of failure function
Fi gur e 3. 3a: Fai l ur e f unct i on of exponent i al di st r i but i on f or di f f er ent val ues
of λ
Failure Function
Increasing function Probability of
failure by given
age
Fraction of
items that fail by
given age
0
0.2
0.4
0.6
0.8
1
0 100 200 300 400
Time
λ · 0.03
F
a
i
l
u
r
e
F
u
n
c
t
i
o
n
,
F
(
t
)
λ · 0.02
λ · 0.01
3. Rel i abi l i t y M easur es 54
Fi gur e 3.3b Fai l ur e f unct i on of Wei bul l di st r i but i on f or di f f er en t β val ues
Fi gur e 3. 3c Fai l ur e f unct i on of nor mal di st r i but i on f or di f f er ent µ val ues
Exampl e 3.1
The t i me t o f ai l ur e di st r i but i on of a sub syst em i n an ai r cr af t engi n e f ol l ows
Wei bul l di st ri but i on w i t h scal e par amet er η = 1100 f l i ght hour s and t he
shape par amet er β = 3. Fi nd:
a) Probability of failure during first 100 flight hours.
b) Find the maximum length of flight such that the failure probability is less
than 0.05.
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200
Ti me
µ = 100
µ = 120
µ = 140
F
a
i
l
u
r
e
F
u
n
c
t
i
o
n
,
F
(
t
)
F
a
i
l
u
r
e
F
u
n
c
t
i
o
n
,
F
(
t
)
β
= 1
β
= 2
β
= 3
Time
3. Rel i abi l i t y M easur es 55
SOLUTION:
a) The f ai l ur e f unct i on f or Wei bul l di st ri but i on is gi ven by:
F t
t
( ) exp( ( ) ) · − −
−
1
γ
η
β
It i s gi ven t hat : t = 100 f l i ght hour s, η = 1100 f l i ght hour s, β = 3 and γ = 0.
Pr obabi l i t y of f ail ur e wi t hi n f i r st 100 hour s is gi ven by:
F( ) exp( ( ) ) . 100 1
100 0
1100
0 00075
3
· − −
−
·
b) If t i s t he maxi mum l engt h of f l i ght such t hat t he f ai l ur e pr obabi l i t y i s l ess
t han 0.05, w e have
3 / 1 3
3
3
)] 95 . 0 ln( [ 1100 95 . 0 ln )
1100
(
95 . 0 ) )
1100
( exp(
05 . 0 ) )
1100
0
( exp( 1 ) (
− × · ⇒ − > ·
> − ·
<
−
− − ·
t
t
t
t
t F
Now sol vi ng f or t , w e get t = 408.70 f l i ght hour s. The maxi mum l en gt h of
f l i ght such t hat t he f ai l ur e pr obabi l i t y i s l ess t han 0. 05 i s 408.70 f li ght hour s.
Exampl e 3.2
The t i me t o f ai l ur e di st r i but i on of a Radar War ni ng Recei ver (RWR) syst em
i n a f i ght er ai r cr af t f ol l ow s Wei bul l di st r i but i on wi t h scal e par amet er 1200
f l i ght hour s and shap e par amet er 3. Th e t i me t o f ai l ur e di st r i but i on of t he
same RWR i n a hel i copt er f ol l ow s exponent i al di st r i but i on w i t h scal e
par amet er 0.001. Compar e t he f ai l ur e f unct i on of t he RWR i n t h e f i ght er
ai r cr af t and t he h el i copt er . If t he suppl i er gi ves a w ar r ant y f or 750 f l i ght
hour s, cal cul at e t he r i sk i nvol ved wi t h r esp ect t o f i ght er ai r cr af t and t he
hel i copt er . (Al t hough we have a same syst em, t he oper at i ng condi t i ons
have si gni f i cant i mpact on t he f ai l ur e f unct i on. In t hi s case, RWR i n
hel i copt er i s subj ect t o mor e vi br at i ons compar ed t o ai r cr af t ).
3. Rel i abi l i t y M easur es 56
SOLUTION:
The f ai l ur e f unct i on of RWR on t he f i ght er ai r cr af t i s given by:
F t
t
( ) exp( ( ) ) · − − 1
1200
3
The f ai l ur e f unct i on of RWR on t he hel i copt er i s gi ven by:
F t t ( ) exp( ( . )) · − − × 1 0001
Fi gur e 3.4 d epi ct s t he f ai l ur e f unct i on of RWR i n f i ght er ai r cr af t and t he
hel i copt er .
3. Rel i abi l i t y M easur es 57
Fi gur e 3.4 Fai l ur e f unct i on of RWR i n f i ght er ai r cr af t and hel i copt er
If t he suppl i er pr ovi des w ar r ant y f or 750 f l i ght hour s t he r i sk associ at ed
wi t h ai r cr af t i s gi ven by:
F( ) exp( ( ) ) . 750 1
750
1200
0 2166
3
· − − ·
That i s, j ust above 21% per cent of RWR ar e l i kel y t o f ai l i f t he RWR i s
i nst al l ed i n t he ai r cr af t .
If t he RWR i s i nst al l ed i n hel i copt er t hen t he associ at ed r i sk i s gi ven by:
F( ) exp( . ) . 750 1 0001 750 05276 · − − × ·
In t he case of hel i copt er , mor e t han 52% of t he RWR’ s ar e l i kel y t o f ai l
bef or e t he war r ant y per i od.
3.12.1 Failure function of system under multiple failure
mechanisms
It i s sel dom t r ue t hat an i t em' s f ai l ur e i s caused by a si ngl e f ai l ur e
mechani sm. In most of t he cases t her e wi l l be mor e t han one (some t i mes
F
a
i
l
u
r
e
F
u
n
c
t
i
o
n
,
F
(
t
)
Hel i copt er
Ai r cr af t
Ti me
0 1000 2000
3000 4000
0.2
0.4
0. 6
0.8
1
0
3. Rel i abi l i t y M easur es 58
hundr eds) mechani sm t hat causes t h e f ai l ur e of an i t em. The expr essi on
(3.1) i s mor e appr opr i at e w hen t he f ai l ur e i s caused by a si ngl e f ai l ur e
mechani sm. However , most of t he pr act i cal syst ems f ai l due t o di f f er en t
causes or di f f er ent f ai l ur e mechani sms. Assume t hat t he syst em f ai l ur e i s
due t o t w o di f f er ent f ai l ur e mechani sms. Let f
1
(t ) and f
2
(t ) be t h e pr obabi l i t y
densi t y f unct i on of t he syst em due t o f ai l ur e mechani sm 1 and 2
r esp ect i vel y. Now t he pr obabi l i t y densi t y f unct i on of t he t i met of ai l ur e of
t he syst em caused by ei t her of t he f ai l ur e mechani sms:
)] ( 1 )[ ( )] ( 1 )[ ( ) (
1 2 2 1
t F t f t F t f t f − + − ·
wher e, F
1
(t ) and F
2
(t ) t he ar e f ai l u r e f unct i on f or f ai l ur e mechani sm 1 and 2
r esp ect i vel y. The f ai l ur e f unct i on of t he i t em und er t wo di f f er en t f ai l ur e
mechani sm i s gi ven by:
∫
− + − ·
t
dx x F x f x F x f t F
0
1 2 2 1
)]} ( 1 )[ ( )] ( 1 )[ ( { ) ( (3.2)
Exampl e 3.3
Fai l ur e of an i t em i s caused by t w o di f f er ent f ai l ur e mechani sms (say f ai l ur e
mechani sm A and B). The t i me t of ai l ur e di st r i but i on of t he i t em due t o
f ai l ur e mechani sm A can be r epr esent ed by exponent i al di st r i but i on wi t h
par amet er λ
A
= 0.002 hour s. The t i met of ai l ur e di st r i but i on of t he i t em due
t o f ai l ur e mechani sm B can be r epr esent ed by exponent i al di st r i but i on w i t h
par amet er λ
B
= 0. 005 hour s. Fi nd t he pr obabi l i t y t hat t he i t em wi l l f ai l
bef or e 500 hour s of oper at i on.
SOLUTION:
Assume t hat f
A
(t ) and f
B
(t ) r epr esent pr obabi l i t y densi t y f unct i on of t he
t i met of ai l ur e r andom var i abl e due t o f ai l ur e mechani sm A and B
r esp ect i vel y. Thus,
f t t F t t
f t t F t t
A A A A A
B B B B B
( ) exp( ) , ( ) exp( )
( ) exp( ) , ( ) exp( )
· − − · −
· − − · −
λ λ λ
λ λ λ
1
1
Now t he f ai l ur e f unct i on of t he i t em i s gi ven by:
3. Rel i abi l i t y M easur es 59
] ) ( exp( 1 [
] ) ( exp( 1 )[ (
] ) ( exp( 1 )[ (
) ) ( exp( ) ) ( exp( { ) (
0
t
t
t
dx x x t F
B A
B A B A B
B A B A A
B A
t
B B A A
λ λ
λ λ λ λ λ
λ λ λ λ λ
λ λ λ λ λ λ
+ − − ·
+ − − + +
+ − − + ·
+
∫
− + + − ·
Fi gur e 3.5 r epr esent s t he f ai l ur e f unct i on due t o f ai l ur e mechani sm 1, 2 and
t he syst em f ai l ur e f unct i on. The pr obabi l i t y t hat t he i t em wi l l f ai l by 500
hour s i s gi ven by:
9698 . 0 )) 500 ) 002 . 0 005 . 0 (( exp( 1 ) 500 ( · × + − − · F
Fi gur e 3.5 Fai l ur e f unct i on due t o di f f er ent f ai l ur e mechani sms
3.13. RELIABILITY FUNCTION
Rel i abi l i t y i s t he abi l i t y of t h e i t em t o mai nt ai n t he r equi r ed f unct i on f or a
speci f i ed per i od of t i me (or mi ssi on t i me) under gi ven oper at i ng co ndi t i ons.
Rel i abi l i t y f unct i o n, R(t ), i s def i ned as t he pr ob abi l i t y t hat t he syst em wi l l not
f ai l dur i ng t he st at ed per i od of t i me, t , under st at ed oper at i ng condi t i ons.
If TTF r epr esen t s t he t i met of ai l ur e r andom var i abl e w i t h f ai l ur e f unct i on
(cumul at i ve di st r i but i on f unct i on) F(t ), t hen t h e r el i abi l i t y f unct i on R(t ) i s
gi ven by:
R(t) = P{the system doesn't fail during [0 , t]} = 1  F(t) (3.3)
0
0. 2
0. 4
0. 6
0. 8
1
0 1000 2000 3000
Ti me
F
a
i
l
u
r
e
F
u
n
c
t
i
o
n
,
F
(
t
)
Fai l ur e mechani sm A
Fai l ur e mechani sm B
Syst em
3. Rel i abi l i t y M easur es 60
In equat i on (3.3) we assume t hat t he age of t he syst em bef or e t he st ar t of
t he mi ssi on i s zer o. Thus t h e equat i on (3.3) i s val i d on l y f or new syst ems or
t hose syst ems who se f ai l ur es ar e not age r el at ed (t hat i s, t he t i met of ai l ur e
f ol l ow s exponent i al di st r i but i on due t o memor y l ess pr oper t y of
expon ent i al di st r i but i on). However , i n most of t he cases t hi s assumpt i on
may not be val i d. I f t he syst em age i s gr eat er t han zer o at t he begi nni ng of
t he mi ssi on, t hen w e have t o cal cul at e mi ssi on r el i abi l i t y f unct i on, w hi ch
wi l l be di scussed l at er . Fi gur e 3.6 depi ct s t he r el at i o n bet ween r el i abi l i t y
f unct i on and t he TTF densi t y f unct i on. R(t ) i s t he ar ea under TTF d ensit y
bet w een t and ∞.
Fi gur e 3.6 Rel i abi l i t y f unct i on of a hypot het i cal pr obabil i t y di st r i but i on
Propert i es of rel i abi l i t y f unct i on:
1. Reliability is a decreasing function with time t. That is, for t
1
< t
2
; R(t
1
)
≥ R(t
2
).
2. It is usually assumed that R (0) = 1. As t becomes larger and larger R(t)
approaches zero, that is, R(∞).
Appl i cat i ons of rel i abi l i t y f unct i on
1. R(t) is the probability that an individual item survives up to time t.
2. R(t) is the fraction of items in a population that survive up to time t.
3. R(t) is the basic function used for many reliability measures and system
reliability prediction.
Rel i abi l i t y f unct i on f or some i mpor t ant l i f e di st r i but i ons ar e gi ven i n Tabl e
3.2. Fi gur e 3.7ac r epr esent s r el i abi l i t y f unct i on of var i ous t heor et i cal
di st r i but i ons f or di f f er ent par amet er val ues.
Ti me
f t ( )
T
T
F
F
a
i
l
u
r
e
d
e
n
s
i
t
y
R(t )
3. Rel i abi l i t y M easur es 61
Table 3.2. Reliability function, R(t), for popular theoretical distributions
Distribution
Reliability function, R(t)
Exponential
0 , 0 ) exp( > > − λ λ t t
Normal
dx e
t
t
x
∫
− ·
−
Φ
,
_
¸
¸
,
_
¸
¸ −
−
0
2
1
2
2
1
1 ) (
σ
µ
π σ
σ
µ
or NORM DIST (µ, t , σ, TRUE) i n EXCEL
Lognormal
dx e
x
t
t
x
l
l
l
l
l
∫
− ·
−
Φ
,
_
¸
¸
,
_
¸
¸ −
−
0
) ln(
2
1
2
2
1
1 )
ln
(
σ
µ
π σ
σ
µ
or NORM DIST (µ, l n(t ), σ, TRUE) i n EXCEL
Weibull
γ γ β η
η
γ
β
≥ >
−
− t
t
, 0 , , ) ) ( exp(
Gamma
∫
Γ
−
− −
t
x
dx e x
0
1
) (
1
1
β α α
β
α
0
0.2
0.4
0.6
0.8
1
0 1000 2000 3000
Ti me
λ
λ·0.001
λ·0.005
λ·0.008 =
R
e
l
i
a
b
i
l
i
t
y
F
u
n
c
t
i
o
n
,
R
(
t
)
3. Rel i abi l i t y M easur es 62
Figure 3.7 a. Reliability function of exponential distribution for different
values of λ
3. Rel i abi l i t y M easur es 63
Fi gur e 3.7 b. Rel i abi l i t y f unct i on of Wei bul l di st r i but i on f or di f f er ent val ues
of β
Fi gur e 3.7c. Rel i abi l i t y f unct i on of Nor mal di st ri but i on f or di f f er ent val ues of
µ
Exampl e 3.4
Ti me t o f ai l ur e di st r i but i on of a comput er memor y chi p f ol l ows nor mal
di st r i but i on w i t h mean 9000 hour s and st andar d devi at i on 2000 hour s.
Fi nd t he r el i abi li t y of t hi s chi p f or a mi ssi on of 8000 hour s.
SOLUTION
Usi ng Tabl e 3.2, t he r el i abil i t y f or a mi ssi on of 8000 hour s i s gi ven by:
R t
t
( ) ( ) ( ) ( . ) . ·
−
·
−
· · Φ Φ Φ
µ
σ
9000 8000
2000
05 0 6915
0
0.2
0.4
0.6
0. 8
1
0 50 100 150
Ti me
R
e
l
i
a
b
i
l
i
t
y
F
u
n
c
t
i
o
n
,
R
(
t
)
µ·50
µ·60
µ·70
0
0.2
0.4
0.6
0. 8
1
0 50 100 150 200
250
Ti me
R
e
l
i
a
b
i
l
i
t
y
F
u
n
c
t
i
o
n
,
R
(
t
)
β·3
β·2
β·1
3. Rel i abi l i t y M easur es 64
Exampl e 3.5
The t i me t o f ai l ur e di st r i but i on of a st eam t ur bo gen er at or can be
r epr esen t ed usi ng Wei bul l di st r i but i on wi t h η = 500 hour s and β = 2.1. Fi nd
t he r el i abi l i t y of t he gener at or f or 600 hour s of oper at i on.
SOLUTION:
Agai n usi ng Tabl e 3.2, r el i abi l i t y of t he gen er at or f or 600 hour s of
oper at i ons i s gi ven by:
R t ( ) exp( ( / ) ) . · − · 600 500 0 2307
2.1
3.13.1 Reliability function for items under multiple failure
mechanisms
Assume t hat t he f ai l ur e of t he i t em i s caused due t o t wo di f f er ent f ai l ur e
mechani sms. Let f
1
(t ) and f
2
(t ) be t he pr obabi l i t y densi t y f unct i on of t he
t i met of ai l ur e r andom var i abl e due t o f ai l ur e mechani sm 1 and 2
r esp ect i vel y. The pr obabi l i t y densi t y f unct i on of t he t i met of ai l ur e of t he
i t em i s gi ven by caused by ei t her of t he f ai l ur e mechan i sms:
)] ( 1 [ ) ( )] ( 1 )[ ( ) (
1 2 2 1
t F t f t F t f t f − ⋅ + − ·
Wh er e F
1
(t ) and F
2
(t ) ar e f ai l ur e f unct i on f or f ai l ur e mechani sm 1 and 2
r esp ect i vel y. Th e Rel i ab i l i t y f unct i on of t h e i t em und er t wo di f f er ent f ai l ur e
mechani sm i s gi ven by:
∫
− + − − · − ·
t
dx x F x f x F x f t F t R
0
1 2 2 1
)]} ( 1 )[ ( )] ( 1 )[ ( { 1 ) ( 1 ) ( (3.4)
The abo ve r esul t can be ext ended t o obt ai n expr essi on f or r el i abi l i t y
f unct i on due t o mor e t han t wo f ai l ur e mechani sms.
Exampl e 3.6
For t he exampl e 3.3, f i nd t he r el i abil i t y of t he i t em f or 200 hour s.
SOLUTION:
3. Rel i abi l i t y M easur es 65
Usi ng t he expr essi on f or f ai l ur e f unct i on obt ai ned i n exampl e 3.3, t he
r el i abi l i t y f unct i on can be w r i t t en as:
2465 . 0 ) 200 ) 005 . 0 002 . 0 ( exp( ) 200 (
) ) ( exp( ) (
· × + − ·
× + − ·
R
t t R
B A
λ λ
3.13.2 Mission Reliability Function
In many pr act i cal si t uat i ons, one mi ght be i nt er est ed i n f i ndi ng t he
pr obabi l i t y of compl et i ng a mi ssi on successf ul l y. Success pr obabi l i t y of
hi t t i ng an en emy t ar get and r et ur ni ng t o t he base i s an exampl e wher e
mi ssi on r el i abi l i t y f unct i on can be used. The mai n di f f er ence bet ween
r el i abi l i t y f unct i on and t he mi ssi on r el i abi l i t y f unct i on i s t hat , i n mi ssi on
r el i abi l i t y we r ecogni se t h e age of t he syst em bef or e t he mi ssi on. M i ssi on
r el i abi li t y i s def i ned, as t he pr obabi l i t y t hat t he syst em aged t
b
i s abl e t o
compl et e mi ssi on dur at i on of t
m
successf ul l y. We assume t hat no
mai nt enance i s per f or m ed dur i ng t he mi ssi on. The expr essi on f or mi ssi on
r el i abi l i t y MR ( t
b
, t
m
) i s gi ven by
) (
) (
) , (
b
m b
m b
t R
t t R
t t MR
+
· (3.5)
wher e, t
b
i s t he age of t he i t em at t he b egi nni ng of t he mi ssi on and t
m
i s t he
mi ssi on per i od. If t he t i me t o f ai l ur e di st r i but i on i s exponent i al , t hen t he
f ol l ow i ng r el at i on i s val i d.
) ( ) , (
m m b
t R t t MR ·
Appl i cat i on of mi ssi on reli abi l i t y f unct i on
1. Mission reliability, MR(t
a
, t
m
) gives the probability that an individual
item aged t
a
will complete a mission duration of t
m
hours without any
need for maintenance.
2. Mission reliability is the appropriate basic reliability measure for ageing
items or items whose timetofailure distribution is other then
exponential.
Exampl e 3.7
3. Rel i abi l i t y M easur es 66
Ti me t of ai l ur e di st r i but i on of t he gear box wi t hi n an ar mour ed vehi cl e can
be model l ed usi ng Wei bul l di st r i but i on wi t h scal e par amet er η = 2400 mi l es
and shape par amet er β = 1.25. Fi nd t h e pr obabi l i t y t hat t hat gear box wi l l
not f ai l dur i ng a mi ssi on t i me of 200 mi l es. Assumi ng t hat t he age of t he
gear box i s 1500 mi l es.
SOLUTION:
Gi ven, t
b
= 1500 mi l es and t
m
= 200 mi l es
) 1500 (
) 1700 (
) (
) (
) , (
R
R
t R
t t R
t t MR
b
b m
m b
·
+
·
5221 . 0 ) )
2400
1700
( exp( ) 1700 (
25 . 1
· − · R
5736 . 0 ) )
2400
1500
( exp( ) 1500 (
25 . 1
· − · R
9102 . 0
5736 . 0
5221 . 0
) 1500 (
) 1700 (
) 200 , 1500 ( · · ·
R
R
MR
That i s, t he gear box aged 1500 mi l es has appr oxi mat el y 91% chance of
sur vi vi ng a mi ssi on of 200 mi l es.
3.14. DESPATCH RELIABILITY
Despat ch r el i abi l i t y (DR) i s one of popul ar r el i abi l i t y met r i cs used by
commer ci al ai r l i nes ar ound t he wor l d. Despat ch r el i abi l i t y i s def i ned as t he
per cent age of r evenu e depar t ur es t hat do not occur i n a del ay or
cancel l at i on due t o t echni cal pr obl ems. For most ai r l i nes, t he d el ay means
t hat t h e ai r cr af t i s del ayed mor e t han 15 mi nut es. Techni cal del ays occur
can be caused due t o some unschedul ed mai n t enan ce. Ai r l i nes f r equent l y
seek DR guar ant ees wher e t h e ai r cr af t manuf act ur es f ace penal t i es i f DR
l evel s ar e not achi eved. For commer ci al ai r l i nes despat ch r el i abi l i t y i s an
i mpor t ant economi c f act or , i t i s est i mat ed t hat del ay cost per mi nut e f or
3. Rel i abi l i t y M easur es 67
l ar ge j et s can be as hi gh as 1000 US dol l ar s. The expr essi on f or despat ch
r el i abi l i t y i s gi ven by:
% 100
100
100
(%)
15
×
− −
·
NC ND
DR (3.6)
Wher e,
ND
1 5
= Number of del ays w i t h mor e t han 15 mi nut es del ay
NC = t he number of cancel l at i ons
Equat i on (3. 6) i s appl i ed onl y t o t echni cal del ays. DR i s a f unct i on of
equi pment r el i abi l i t y, syst em and comp onent mai nt ai nabi l i t y, and over al l
l ogi st i c suppor t .
3.15. HAZARD FUNCTION (HAZARD RATE OR
INSTANTANEOUS FAILURE RATE)
Hazar d f unct i on (or hazar d r at e) i s used as a par amet er f or compar i son of
t wo di f f er en t desi gns i n r el i abi l i t y t heor y. Hazar d f unct i on i s t he i ndi cat or of
t he ef f ect of agei ng on t he r el i abi l i t y of t he syst em. It quant i f i es t he r i sk of
f ai l ur e as t he age of t h e syst em i ncr eases. M at hemat i cal l y, i t r epr esent s t he
condi t i onal pr obabi l i t y of f ai l ur e i n an i nt er val t t o t + δt gi ven t hat t he
syst em sur vi ves up t o t , di vi ded by δt , as δt t ends t o zer o , t hat i s,
) (
) ( ) (
lim
) (
) ( ) (
.
1
lim ) (
0 0 t tR
t t R t R
t R
t F t t F
t
t h
t t δ
δ δ
δ δ δ
+ −
·
− +
·
→ →
(3.7)
Not e t hat hazar d f unct i on, h(t ), i s not a pr obabi l i t y, i t i s t he l i mi t i ng val ue of
t he pr obabi l i t y. How ever , h(t )δt , r epr esent s t h e pr o babi l i t y t hat t he i t em
w i l l f ai l bet w een ages t and t +δt as δt →0. The abo ve expr essi on can be
si mpl i f i ed so t hat
h t
f t
R t
( )
( )
( )
· (3.8)
3. Rel i abi l i t y M easur es 68
Thus, t he hazar d f unct i on i s t he r at i o of t he pr obabi l it y densi t y f unct i on t o
t he r el i abi l i t y f unct i on. Int egr at i ng bot h si des of t he above equat i on, we
get :
h x dx
f x
R x
dx
R x
R x
dx R t
t t
t
( )
( )
( )
' ( )
( )
ln ( )
0 0
0
∫
·
∫
· −
∫
· −
Thus r el i abi l i t y can be w r i t t en as:
R t h x dx
t
( ) exp ( ) · −
¸
1
]
1
1
∫
0
(3.9)
Fr om equat i on (3.9), i t i mmedi at el y f ol l ow s t hat :
f t h t h x dx
t
( ) ( ) exp( ( ) · −
∫
0
(3.10)
The expr essi on (3.10), whi ch r el at es r el i abi l i t y and hazar d f unct i on, i s val i d
f or al l t ypes of t i me t o f ai l ur e di st r i but i on. Hazar d f unct i on shows how t he
r i sk of t he i t em i n use changes over t i m e (h ence al so cal l ed r i sk r at e). The
hazar d f unct i ons of so me i mpor t ant t h eor et i cal di st r i but i ons ar e gi ven i n
Tabl e 3.3.
Charact eri st i cs of hazard f unct i on
1. Hazard function can be increasing, decreasing or constant.
2. Hazard function is not a probability and hence can be greater than 1.
Table 3.3. Hazard function, h(t), of few theoretical distributions
Distribution Hazard function, h(t)
Exponential λ
3. Rel i abi l i t y M easur es 69
Distribution Hazard function, h(t)
Normal
), ( / ) (
σ
µ t
t f
−
Φ f(t) is the pdf of normal distribution.
Lognormal
), ( / ) (
l
l
l
t
t f
σ
µ −
Φ f
l
(t) is the pdf of lognormal
distribution.
Weibull
1
) (
− β
η η
β t
Gamma
[
( )
] /
( )
β
α α
β
α
α β α α β
Γ Γ
t e x e dx
t x
t
− − − −
−
∫
1 1
0
1
1
Appl i cat i ons of hazard f unct i on
1. h(t) is loosely considered as failure rate at time t (timedependent)
2. h(t) quantifies the amount of risk a system is under at time t.
3. For h(t) ≤ 1, it is not recommended to carry out preventive maintenance.
Fi gur es 3.8ac show hazar d f unct i on of var i ous t heor et i cal di st r i but i ons f or
di f f er en t par amet er val ues.
0
0.02
0.04
0.06
0.08
0 50 100 150 200 250
Time
h
a
z
a
r
d
f
u
n
c
t
i
o
n
β = 2.4
β = 0.8
β = 1.6
3. Rel i abi l i t y M easur es 70
Fi gur e 3.8a Hazar d f unct i on of Wei bul l di st r i but i on f or di f f er ent val ues of β
Figure 3.8b Hazard function of exponential distribution
Fi gur e 3.8c Hazar d f unct i on of nor mal di st r i but i on f or di f f er ent val ues of µ
Exampl e 3.8
Ti me t o f ai l ur e di st r i but i on of a gas t ur bi n e syst em can be r epr esent ed
usi ng Wei bul l di st r i but i on wi t h scal e par amet er η = 1000 hour s and shape
par amet er β = 1. 7. Fi nd t h e hazar d r at e of t he gas t ur bi ne at t i me t = 800
hour s and t = 1200 hour s.
SOLUTION:
The hazar d r at e f or Wei bul l di st r i but i on i s gi ven by:
0
0. 005
0. 01
0. 015
0. 02
0 50 100 150 200
Ti me
h
a
z
a
r
d
f
u
n
c
t
i
o
n
0
0. 1
0. 2
0. 3
0. 4
0. 5
0. 6
0 20 40 6 0 80 100
Ti m e
h
a
z
a
r
d
f
u
n
c
t
i
o
n
µ = 50
µ = 60
µ = 70
3. Rel i abi l i t y M easur es 71
h t
t
( ) ( ) ·
−
β
η η
β 1
h(800 ) =
17
1000
800
1000
0 00145
0 7
.
( ) .
.
·
h(1200 ) =
17
1000
1200
1000
0 0019
0 7
.
( ) .
.
·
3.15.1 Cumulative hazard function
Cumul at i ve hazar d f unct i on r epr esent s t he cumul at i ve hazar d or r i sk of t he
i t em dur i ng t h e i nt er val [ 0,t ] . Cumul at i ve hazar d f unct i on, H(t ), i s gi ven by:
H t h x dx
t
( ) ( ) ·
∫
0
(3.11)
Rel i abi l i t y of an i t em can b e conveni ent l y wr i t t en usi ng cumul at i ve hazar d
as:
R t e
H t
( )
( )
·
−
(3.12)
3.15.2 Cumulative hazard function and the expected
number of failures
Consi der an i t em, whi ch upon f ai l ur e i s subj ect t o mi ni mal r epai r . That i s,
t he hazar d r at e af t er r epai r i s same as t he hazar d r at e j ust bef or e f ai l ur e. I f
N(t ) i s t he t ot al number of f ai l ur es by t i me t , t hen M (t ) = E [ N(t )] i s t he
expect ed numb er of f ai l ur es by t i me t . It can b e show n t hat und er t he
assumpt i on t hat t h e i t em r ecei ves mi ni mal r epai r
*
(‘ asbad asol d’ ) af t er
each f ai l ur e, t hen
E[N(t)] =
∫
·
t
dx x h t M
0
) ( ) ( (3.13)
*
Mathematically minimal repair or as bad as old means that the hazard rate of the item
after repair will be same as the hazard rate just prior to failure.
3. Rel i abi l i t y M easur es 72
The abo ve expr essi on can be used t o model di f f er en t
mai nt enan ce/ r epl acemen t pol i ci es. I n case of exponent i al and Wei bul l t i me
t o f ai l ur e di st r i but i ons we get t he f ol l ow i ng si mpl e expr essi ons f or t he
exp ect ed number of f ai l ur es of an i t em subj ect t o mi ni mal r epai r .
Exponent i al t i me t o f ai l ure di st ri but i on
For exponen t i al di st ri but i on, t he expect ed number of f ai l ur es i s gi ven by
E[N(t)] = h x dx dx t
t t
( )
0 0
∫
·
∫
· λ λ (3.14)
Wei bul l t i me t o f ai l ure di st ri but i on
E[N(t)] = h x dx
x
dx
t
t t
( ) ( ) ( )
0
1
0
∫ ∫
· ·
−
β
η η η
β β
(3.15)
Exampl e 3.9
An i t em i s subj ect t o mi ni mal r epai r whenever i t f ai l ed. I f t h e t i me t o f ai l ur e
of t he i t em f ol l ow s Wei bul l di st r i but i on w i t h η = 500 and β = 2. Fi n d: 1. The
number of t i mes t he i t em i s expect ed t o f ai l by 1500 hour s, and 2. The cost
of t he i t em i s $ 200. If t he cost of mi nimal r epai r i s $ 100 per each r epai r , i s
i t advi sabl e t o r epai r or r epl ace t he i t em upon f ai l ur e.
SOLUTION:
1. The expect ed numb er of f ai l ur es i s gi ven by:
E[N(t)] = [ ] [ ]
t
η
β
· · ·
1500
500
3 9
2 2
2. Usi ng t he above r esul t t he cost associ at ed w i t h r epai r , C
rep air
(t ) = 9 × 100
= $ 900.
If t he i t em i s r epl aced, t hen t he expect ed number of f ai l ur es i s gi ven by t he
r enewal f unct i on, M (t ) [ r ef er chapt er 4] , w her e
3. Rel i abi l i t y M easur es 73
M t F t
i
i
( ) ( ) ·
·
∞
∑
1
For t he above case, t he val ue of M (t ) < 4 (Th e act ual cal cul at i on of t he
above f unct i on wi l l be di scussed i n Chapt er 4). Thus t he cost due t o
r epl acement wi l l be l ess t han 4 × 200 = $ 800. Thus, i t i s bet t er t o r epl ace
t he i t em upon f ai l ur e r at her usi ng mi ni mal r epai r .
3.15.3 Typical Forms of Hazard Function
In pr act i ce, hazar d f unct i on can have di f f er ent shapes. Fi gur e 3.9 shows
most gen er al f or ms of hazar d f unct i on. Recent r esear ch i n t he f i el d of
r el i abi l i t y cent r ed mai nt enance (RCM ) shows t hat t h e hazar d r at e most l y
f ol l ow s six di f f er ent pat t er ns. Dependi ng on t h e equi pment and i t s f ai l ur e
mechani sm, one can say t hat t he hazar d f unct i on may f ol l ow any one of
t hese si x pat t er ns. How ever , one shoul d not bl i ndl y assume t hat hazar d
r at e of any i t em w i l l f ol l ow any one of t hese si x pat t er ns. These ar e onl y
possi bl e cases based on some dat a.
Pat t er n A i s cal l ed t he bat ht ub cur ve and consi st of t hr ee di st i nct phases. I t
st ar t s wi t h ear l y f ai l ur e r egi on (known as bur ni n or i nf ant mor t al i t y)
char act er i sed by decr easi ng hazar d f unct i on. Ear l y f ai l ur e r egi on i s f ol l ow ed
by const an t or gr adual l y i ncr easi ng r egi on (cal l ed usef ul li f e). The const ant
or gr adual l y i ncr easi ng r egi on i s f ol l owed b y w ear out r egi on char act er i sed
by i ncr easi ng hazar d f unct i on. The r eason f or such as shape i s t hat t he
ear l y decr easi ng hazar d r at e r esul t s f r om manuf act ur i ng def ect s. Ear l y
oper at i on wi l l r emove t h ese i t ems f r o m a popul at i on of l i ke i t ems. The
r emai ni ng i t ems have a const ant hazar d f or some ext ended per i od of t i me
dur i ng whi ch t h e f ai l ur e cause i s not r eadi l y appar ent . Fi nal l y t hose i t ems
r emai ni ng r each a w ear out st age wi t h an i ncr easi ng hazar d r at e. One
woul d expect bat ht ub cur ve at t h e syst em l evel and not at t he par t or
componen t l evel (unl ess t h e compo nent has many f ai l ur e modes whi ch
have di f f er ent TTF di st r i but i on). It was bel i eved t hat bat h t ub cur ve
r epr esen t s t he most gener al f or m of t he hazar d f unct i on. However , t he
r ecent r esear ch shows t hat i n most of t he cases hazar d f unct i on do not
f ol l ow t hi s pat t er n.
Pat t er n B st ar t s wi t h hi gh i nf ant mor t al i t y and t hen f ol l ows a const ant or
ver y sl owl y i ncr easi ng hazar d f unct i on . Pat t er n C st ar t s wi t h a const ant or
sl owl y i ncr easi ng f ai l ur e pr obabi l i t y f ol l owed by wear out (shar pl y
i ncr easi ng) hazar d f unct i on. Pat t er n D shows const ant hazar d t hr oughout
t he f il e. Pat t er n E r epr esent s a sl owl y i ncr easing hazar d w i t hout any si gn of
3. Rel i abi l i t y M easur es 74
wear out . Pat t er n F st ar t s wi t h a l ow hazar d i ni t i al l y f ol l owed by a const ant
hazar d.
Figure 3.9. Different forms of hazard function
Tabl e 3.4 shows t he r el at i onshi p bet ween f ai l ur e f unct i on, r el i abi l i t y
f unct i on and hazar d f unct i on.
Tabl e 3.4. Rel at i onshi p bet w een F(t ), R(t ) and h(t )
F(t) R(t) h(t)
F(t)  1  R(t)
∫
− −
t
dx x h
0
) ) ( exp( 1
Pattern A: Hazard function (bathtub curve)
0
0.01
0.02
0.03
0.04
0 20 40 60 80 100
Age
Pattern B: High infant Mortality
0
0.01
0.02
0.03
0 100 200 300
Age
Pattern D: Constant Hazard
0
0.002
0.004
0.006
0.008
0.01
0.012
0 50 100 150
Age
Pattern E: Slowly increasing hazard
function
0
0.0002
0.0004
0.0006
0 50 100 150
Age
Pattern C: Slowly increasing hazard
followed by wearout
0.00995
0.01
0.01005
0.0101
0 50 100 150
Age
Pattern F: Initial low hazard followed by
constant hazard
0
0.005
0.01
0.015
0 50 100 150 200 250
Age
3. Rel i abi l i t y M easur es 75
F(t) R(t) h(t)
R(t) 1  F(t) 
∫
−
t
dx x h
0
) ( exp(
h(t)
)] ( 1 [ ) (
'
t R t F − ) ( ) (
'
t R t R −

3.15.4 Failure rate
Wh enever t he hazar d f unct i on i s const ant , w e cal l i t as f ai l ur e r at e. That i s,
f ai l u r e r at e i s a speci al case of hazar d f unct i on (w hich i s t i me dep end ent
f ai l ur e r at e). Fai l ur e r at e i s one of t he most wi del y used cont r act ual
r el i abi l i t y measur es i n t he d ef ence and aer ospace i ndust r y. By def i ni t i on, i t
i s appr opr i at e t o use f ai l ur e r at e onl y when t he t i met of ai l ur e di st r i but i on
i s exponent i al . Al so, f ai l ur e r at e can be used onl y f or a nonr epai r abl e
syst em. M any def ence st andar ds such as M ILHDBK 217 and Br i t i sh DEF
STAN 0040 r eco mmend t h e f ol l owi ng equat i on f or est i mat i ng t he f ai l ur e
r at e.
sample the of time operating Cumulative
sample a in failures of number Total
rate Failure · (3.16)
Car e shoul d be t aken i n usi ng t he above equat i on, f or good est i mat i on one
has t o obser ve t he syst em f ai l ur e f or a suf f i ci ent l y l ar ge oper at i ng per i od.
Appl i cat i ons of f ai l ure rat e
1. Failure rate represents the number of failures per unit time.
2. If the failure rate is λ, then the expected number of items that fail in [0,t]
is λt.
3. Failure rate is one of the popular contractual reliability measures among
many industries including aerospace and defence.
3. Rel i abi l i t y M easur es 76
3.16. MEAN TIME TO FAILURE (MTTF)
M TTF r epr esent s t h e exp ect ed val ue of a syst em' s t i me t o f ir st f ai l ur e. It i s
used as a measu r e of r el i abi l i t y f or nonr epai r abl e i t ems such as bul b,
mi cr ochi ps and many el ect r oni c ci r cui t s. M at hemat i cal l y, M TTF can be
def i ned as:
MTTF tf t dt R t dt · ·
∞ ∞
∫ ∫
( ) ( )
0 0
(3.17)
Thus, M TTF can be consi der ed as t h e ar ea under t he cur ve r epr esen t ed
by t he r eli abili t y f unct ion, R(t ), bet w een zer o and i nf i ni t y. If t he i t em under
consi d er at i on i s r epai r abl e, t hen t he expr essi on (3. 17) r epr esent s m ean
t i me t o f i r st f ai l ur e of t he i t em. Fi gur e 3.10 depi ct s t he M TTF val ue of an
i t em.
For many r el i abi l i t y f unct i ons, i t i s di f f i cul t t o eval uat e t h e i nt egr al (3.17).
One may have t o use numer i cal appr oxi mat i on such as t r apezi um appr oach
t o f i nd M TTF val ue.
Fi gur e 3.10 M TTF of an i t em as a f unct i on of Rel i abi l i t y
M TTF i s one of t h e most popul ar measur es f or sp eci f yi ng r el i abi l i t y of non
r epai r abl e i t ems among mi l i t ar y and Gover nment or gani sat i ons t hr oughout
t he wor l d. Unf or t unat el y t h er e ar e many mi sconcept i on about M TTF
among r el i abi l i t y anal yst s. Dur i ng t he Gul f War , on e of Gener al s f r om a
def ence depar t m ent sai d, ' We know exact l y how many t anks t o send, w e
measur ed t he di st ance f r om t he map and di vi ded t hat by M TTF' . What many
peopl e do no t r eal i se i s t hat M TTF i s onl y a measur e of cent r al t enden cy.
0
0.2
0.4
0.6
0.8
1
0
1
5
0
3
0
0
4
5
0
6
0
0
7
5
0
9
0
0
1
0
5
0
1
2
0
0
1
3
5
0
1
5
0
0
1
6
5
0
1
8
0
0
1
9
5
0
2
1
0
0
2
2
5
0
Time
R
(
t
)
MTTF
3. Rel i abi l i t y M easur es 77
For exampl e, i f t he t i met of ai l ur e di st r i but i on i s exponent i al , t hen 63% of
t he i t ems w i ll f ai l bef or e t h ei r age r eaches M TTF val ue.
M TTF i s one of t h e i mpor t ant cont r act ual r el i abi l i t y measur es f or non
r epai r abl e (consu mabl e) i t ems. How ever , i t i s i mpor t ant t o under st and
what M TTF val ue r eal l y means. For exampl e l et us assu me t hat we have t wo
i t ems A and B wi t h sam e M TTF (say 500 days). One mi ght t hi nk t hat bot h
t he co mpon ent s have equal r el i abi l i t y. However , i f t he t i me t o f ai l ur e of
t he i t em A i s expon ent i al i s t hat of i t em B i s nor mal t hen t her e wi l l be a
si gni f i cant var i at i on i n t he behavi our of t hese i t ems. Fi gur e 3.11 sh ows t he
cumul at i ve di st r i but i on of t hese t wo i t ems up t o 500 days. The f i gur e
cl ear l y shows t hat i t ems wi t h exponent i al f ai l ur e t i me show hi gh er chance
of f ail ur e dur i ng t he i ni t i al st ages of oper at i on.
Fi gur e 3.11 Compar i son of i t em w i t h same M TTF
Usi ng t he equat i on (3.17), t he M TTF of var i ous f ai l ur e di st r i but i ons ar e
l i st ed i n Tabl e 3.5.
It i s easy t o check t hat i f t he t i me t o f ai l ur e of t he i t em i s exponent i al t hen
mor e t han 63% of t he i t ems w i l l f ail by t he t i me t he age of t he i t em r eaches
M TTF. I n t he case of nor mal di st r i but i on, i t wi l l be 50%.
Appl i cat i ons of MTTF
1. MTTF is the average life of a nonrepairable system.
2. For a repairable system, MTTF represents the average time before the
first failure.
0
0.2
0.4
0.6
0.8
1
0 200 400 600 800 1000
Time
F
a
i
l
u
r
e
F
u
n
c
t
i
o
n
Normal
Exponential
3. Rel i abi l i t y M easur es 78
3. MTTF is one of the popular contractual reliability measures for non
repairable systems.
Table 3.5. MTTF of different timetofailure distributions
Distribution MTTF
Exponential 1/λ
Normal µ
Lognormal
)
2
exp(
l
l
σ
µ +
Weibull
)
1
1 (
β
η + Γ ×
Gamma α / β
3.16.1 Mean Residual Life
In some cases, i t may be of i nt er est t o know t h e expect ed val ue of t he
r emai ni ng li f e of t he i t em bef or e i t f ai l s f r om an ar bi t r ar y t i me t
0
(know n as,
mean r esi dual l i f e). We denot e t hi s val ue as M TTF( t
0
), w hi ch r epr esent s t he
expect ed t i me t o f ai l ur e of an i t em aged t
0
. M at hemat i cal l y, M TTF(t
0
) can
be expr essed as:
MTTF t t t f t t dt
t
( ) ( ) (  )
0 0 0
0
· −
∞
∫
(3.18)
f (t  t
0
) i s t he d ensi t y of t he condi t i onal pr obabi l i t y of f ai l ur e at t i me t ,
pr ovi ded t hat t he i t em has sur vi ved over t i me t
0
. Thus,
)  ( ) ( )  (
0 0
t t R t h t t f × ·
3. Rel i abi l i t y M easur es 79
wher e, R(t  t
0
), i s t he condi t i onal pr obabi l i t y t hat t he i t em sur vi ves up t o
t i me t , gi ven t hat i t has sur vi ved up t o t i me t
0
. Now , t he above expr essi on
can be w r i t t en as:
) (
) (
) ( )  (
0
0
t R
t R
t h t t f × ·
The expr essi on f or M TTF(t
0
) can be wr i t t en as:
MTTF t t t h t
R t
R t
dt
t
( ) ( ) ( )
( )
( )
0 0
0
0
· −
∞
∫
(3.19)
subst i t ut i ng f or h(t ) i n t he above equat i on, we have
MTTF t
t t f t
R t
dt
R t
t t f t dt
t t
( )
( ) ( )
( ) ( )
( ) ( )
0
0
0 0
0
0 0
1
·
−
· −
∞ ∞
∫ ∫
The above eq uat i on can be w r i t t en as (usi ng i nt egr at i o n by par t s):
MTTF t
R t dt
R t
t
( )
( )
( )
0
0
0
·
∞
∫
(3.20)
The concept of mean r esi dual l i f e can be successf ul l y appl i ed f or pl anni ng
mai nt enance and i nspect i on act i vi t i es.
Exampl e 3.10
Compani es A and B manuf act ur e car t yr es. Bo t h t he compani es cl ai m t hat
t he M TTF of t hei r car t yr e i s 2000 mi l es. Af t er anal ysi ng t he f i el d f ai l ur e
dat a of t h ese t wo t yr es i t was f ound t hat t he t i me t o f ai l ur e di st r i but i on of
A i s exponent i al wi t h λ = 0.0005 and t he t i me t o f ai l ur e di st r i but i on of B i s
nor mal w i t h µ = 2000 mi l es and σ = 200 mi l es. If t he mai nt enan ce pol i cy of
t he Exet er ci t y car r ent al s i s t o r epl ace t he t yr es as soon as i t r eaches 2000
mi l es w hi ch t yr e t hey shoul d buy:
SOLUTION:
3. Rel i abi l i t y M easur es 80
Rel i abi l i t y of t he car t yr e pr oduced by company A f or 2000 mi l es, R
A
(2000),
i s gi ven by:
3678 . 0 ) 2000 0005 . 0 exp( ) 2000 ( · × − ·
A
R
Rel i abi l i t y of t he car t yr e pr oduced by co mpany B f or 2000 mi l es, R
B
(2000),
i s gi ven by:
R
B
( ) ( ) ( ) ( ) . 2000
2000 2000 2000
200
0 05 ·
−
·
−
· · Φ Φ Φ
µ
σ
Thus, i t i s advi sabl e t o buy t he t yr es pr oduced by company B.
Exampl e 3.11
The t i me t o f ai l ur e of an ai r bor ne navi gat i on r adar can be r epr esent ed usi ng
Wei bul l di st ri but i on wi t h scal e par amet er η = 2000 hour s and β = 2.1. I t
was t ol d t hat t he age of t he exi st i ng r adar i s 800 hour s. Fi nd t he expect ed
val ue of t he r emai ni ng l i f e f or t hi s r adar.
SOLUTION:
Usi ng Equat i on (3.20), The M TTF(800) can be w r i t t en as:
) 800 (
) ( ) (
) 800 (
) (
) 800 (
0
800
0 800
R
dt t R dt t R
R
dt t R
MTTF
∫ ∫
−
·
∫
·
∞ ∞
MTTF(800) = MTTF
MTTF
t
dt
( )
exp( ( )
.
.
800
2000
08641
2 1
0
800
·
− −
∫
MTTF · × + · ⋅ + · η
β
Γ Γ ( ) (
.
) . 1
1
2000 1
1
21
17712
The value of )
1
1 (
β
+ Γ can be found from Gamma function table (see
appendix).
3. Rel i abi l i t y M easur es 81
Using numerical approximation,
∫
≈ −
800
0
1 . 2
90 . 763 )
2000
( exp( dt
t
Thus M TTF(800) ≈ (1771.2  763.90 ) / 0.8641 = 1165.72 hour s
Thus, expect ed r emai ni ng l i f e of t he r adar aged 800 hour s i s 1165.72 hour s.
3.16.2 MTTF of a maintained system
Assume t hat an i t em i s subj ect t o pr event i ve mai nt enance af t er ever y T
p m
uni t s, t hat i s, at T
p m
, 2T
p m
, 3T
p m
, et c. The expect ed t i me t o f ai l ur e, M TTF
p m
,
(M TTF of an subj ect t o pr event i ve) of t he i t em i s gi ven by:
MTTF R t dt
pm pm
·
∞
∫
( )
0
(3.21)
Usi ng addi t i ve pr oper t y of i nt egr at i on, t he above i nt egr al can be wr i t t en as:
MTTF R t dt R t dt R t dt
pm pm
T
pm
T
T
pm
T
T
pm
pm
pm
pm
pm
· + + +
∫ ∫ ∫
( ) ( ) ( ) ....
0
2
2
3
wher e R
pm
(t ) i s t he r el i abi l i t y of t he i t em subj ect t o pr event i ve
mai nt enan ce. If t he i t em i s r est or ed t o ‘ as goodas new’ st at e af t er each
mai nt enan ce act i vi t y, t hen t he r el i abi l i t y f unct i on bet w een any t wo
mai nt enan ce t asks can be wr i t t en as:
R t R T R t kT t k T
pm pm
k
pm pm
( ) [ ] ( ), ( ) · ≤ ≤ +1
Usi ng t he above expr essi on f or R
pm
(t ) i n t he i nt egr al (3.21) w e have:
MTTF R t dt R T R t dt R T R t dt
R T R T R t dt
pm
T
pm
T
pm
T
pm pm
T
pm pm pm
pm
· + + +
· + + +
∫ ∫ ∫
∫
( ) ( ) ( ) [ ( )] ( ) ...
{ ( ) [ ( )] ....} ( )
0 0
2
0
2
0
1
3. Rel i abi l i t y M easur es 82
As R(t) ≤ 1, the above expression can be written as:
MTTF
R t dt
R T
R t dt
F T
pm
T
pm
T
pm
pm pm
·
−
·
∫ ∫
( )
( )
( )
( )
0 0
1
(3.22)
Si mi l ar l ogi c can be used t o d er i ve t h e expr essi on f or M TTF
pm
w hen t he
r epai r i s not per f ect (t hat i s, when t he i t em i s not as good as new af t er
mai nt enan ce). M TTF
p m
can be used t o quan t i f y t he ef f ect i veness of t he
mai nt enan ce act i on. If M TTF
pm
>M TTF, t hen one can say t hat t h e r el i abi l i t y
can b e i mpr oved by car r yi ng out mai nt enance. If M TTF
p m
≤ M TTF t h en, t he
mai nt enance w i l l not i mpr ove t he r el i abi l i t y of t he i t em. Fi gur e 3.12 shows
M TTF
p m
val u es of an i t em f or di f f er ent T
p m
w hose t i met of ai l ur e can be
r epr esen t ed usi ng Wei bul l di st r i but i on wi t h η = 200 and β = 2.5. It can be
not i ced t hat as t he val ue of T
pm
i ncr eases, t he M TTF
pm
conver ges t o t hat of
cor r ect i ve mai nt enance.
Exampl e 3.12
A sol i d st at e r adar i s subj ect t o pr even t i ve mai nt enance af t er ever y 400
f l i ght hour s. The t i me t o f ai l ur e of t h e r adar f ol l ows exponent i al
di st r i but i on wi t h mean l i f e 800 f l i ght hour s. Fi n d t he M TTF
p m
of t he r adar .
SOLUTION:
We have: T
0
= 500 f l i ght hour s and (1/ λ) = 800
λ = (1/ 800) = 0.00125
MTTF
t dt
pm
·
− ×
− − ×
·
∫
exp( . )
exp( . )
000125
1 000125 400
800
0
400
Ther e i s no i mpr ovement i n t he M TTF
pm
b ecause t he t i me t o f ai l ur e i s
expon ent i al . Thus, pr event i ve mai nt enance wi l l not i mpr ove t he r el i abi l i t y
of t he syst em, i f t he t i me t o f ai l ur e i s exponent i al . Thi s exampl e i s used t o
demonst r at e t hi s wel l known f act mat hem at i cal l y.
3. Rel i abi l i t y M easur es 83
Fi gur e 3.12. M TTF
pm
of an i t em f or di f f er ent T
p m
val ues
Exampl e 3.13
A manuf act ur i ng company buys t wo machi nes A and B. The t i me t o f ai l ur e
of machi ne A can b e r epr esent ed by Wei bul l di st r i but i on wi t h η = 1000
hour s and β = 2. The t i me t o f ai l ur e of machi n e B can be r epr esent ed by
Wei bul l di st r i but i on w i t h η = 1000 hour s and β = 0.5. Th e mai nt enance
manager i n char ge of op er at i on pl an t o appl y pr event i ve mai n t enance f or
bot h t h e machi n es f or ever y 200 hour s, so t hat he can i mpr o ve t he
expect ed t i me t o f ai l ur e of t he machi nes. Ch eck w het h er t h e manager ’ s
deci si on i s cor r ect .
SOLUTION:
The M TTF
pm
f or machi ne A i s gi ven by:
MTTF
pm
= MTTF
t dt
pm
exp( ( / ) )
exp( ( / ) )
−
−
≈
∫
1000
200 1000
5033
2
0
200
2
hours
MTTF for machine A is η × + · × + · Γ Γ ( ) ( ) . 1
1
1000 1
1
2
886 2
B
hours
Thus f or machi ne A, pr event i ve mai nt enance w i l l i mpr ove t he mean t i me t o
f ai l ur e of t he syst em.
The M TTF
pm
f or machi ne B i s gi ven by:
MTTFpm for different Tpm values
0
500
1000
1500
2000
2500
40 80 120 160 200 240
Tpm
M
T
T
F
p
m
MTTFpm
3. Rel i abi l i t y M easur es 84
MTTF
t dt
pm
·
−
−
≈
∫
exp( ( / ) )
exp( ( / ) )
.
.
1000
200 1000
414
0 5
0
200
0 5
hours
MTTF for machine B is η × + · × + · Γ Γ ( ) (
.
) 1
1
1000 1
1
05
2000
B
hours
Thus f or machi ne B, pr event i ve mai nt enance wi l l decr ease t he mean t i me
t o f ai l ur e of t h e syst em. Thus, i t i s bet t er not t o appl y pr event i ve
mai nt enance f or machi ne B.
3.16.3 Variance of Mean Time To Failure
It i s i mpor t ant t o know t he var i ance of mean t i m e t o f ai l ur e f or bet t er
under st andi ng of t he i t em. Fr om def i ni t i on var i ance V(t ) i s gi ven by:
V t E t E t
t f t MTTF
( ) ( ) [ ( )]
( )
· −
· −
∞
∫
2 2
2
0
2
Appl yi ng i nt egr at i on by par t s:
V t tR t dt MTTF ( ) ( ) · −
∞
∫
2
0
2
(3.23)
3.17. MEAN OPERATING TIME BETWEEN
FAILURES (MTBF)
M TBF st ands f or mean oper at i ng t i me bet ween f ai l ur es (wr ongl y ment i oned
as mean t i me bet ween f ai l ur es t hr oughout t h e l i t er at ur e) and i s used as a
r el i abi l i t y measur e f or r epai r abl e syst ems. In Br i t i sh St andar d (BS 3527)
M TBF i s def i ned as f ol l ow s:
For a st at ed per i od i n t he l i f e of a f unct i onal uni t , t he mean val ue of t he
l engt hs of t i me bet ween consecut i ve f ai l ur es under st at ed condi t i on.
3. Rel i abi l i t y M easur es 85
M TBF i s ext r em el y di f f i cul t t o pr edi ct f or f ai r l y r el i abl e i t ems. However , i t
can be est i mat ed i f t he appr opr i at e f ai l ur e dat a i s avai l abl e. I n f act , i t i s ver y
r ar el y pr edi ct ed w i t h an accept abl e accur acy. In 1987 t h e US Ar my
conduct ed a sur vey of t he pur chase of t hei r SINCGARS r adi os t hat had been
subj ect ed t o comp et i t i ve pr ocur ement and del i ver y f r o m 9 di f f er en t
suppl i er s. Th ey w ant ed t o est abl i sh how t he obser ved Rel i abi l i t y Inser vi ce
compar ed t o t h at whi ch had been pr edi ct ed by each suppl i er (usi ng M IL
HDBK217). Th e ou t put of t hi s exer ci se i s shown i n Tabl e 3.6 ( Knowl es,
1995). It i s i nt er est i ng t o not e t hat t hey ar e al l same r adi o, same d esi gn,
same choi ce of component s (bu t di f f er ent manuf act ur er s) and t he
r equi r ement set b y t he Ar my w as M TBF of 1250 hour s w i t h a 80%
conf i dence. M aj or i t y of t he suppl i er s' obser ved M TBF was no wher e near
t hei r pr edi ct i on.
Tabl e 3.6 SINCGARS r adi os 217 pr edi ct i on and t he obser ved M TBF
Vendor M ILHDBK217 (hour s) Obser ved M TBF (hour s)
A 7247 1160
B 5765 74
C 3500 624
D 2500 2174
E 2500 51
F 2000 1056
G 1600 3612
H 1400 98
I 1000 472
Let us assume t hat t he sequence of r andom var i abl es X
1
, X
2
, X
3
, …X
n
r epr esen t t h e o per at i ng t i me of t h e i t em bef or e i  t h f ai l ur e (Fi gur e 3.13).
M TBF can b e pr edi ct ed by t aki ng t he aver age of expect ed val ues of t he
r andom var i abl es X
1
, X
2
, X
3
,…, X
n
et c. To det er mi ne t hese exp ect ed val ues
i t i s necessar y t o det er mi ne t h e di st r i but i on t ype and par amet er s. As soon
as an i t em f ai l s, appr opr i at e mai nt enan ce act i vi t i es wi l l be car r i ed out . Thi s
i nvol ves r epl aci ng t he r ej ect ed component s wi t h ei t her new on es or ones
t hat have been pr evi ousl y r eco ver ed (r epai r ed). Each of t hese compon ent s
wi l l have a di f f er en t wear out char act er i st i c gover ned by a di f f er ent
di st r i but i on. To f i nd t he expect ed val ue of t he r an dom var i abl e X
2
one
shoul d t ake i nt o account t he f act t hat not al l component s of t he i t em ar e
new and, i ndeed, t hose, whi ch ar e not new, may have qui t e di f f er ent ages.
Thi s makes i t al most i mpossi bl e t o det er mi ne t h e di st r i but i on of t he
r andom var i abl e X
2
and hence t he expect ed val ue.
3. Rel i abi l i t y M easur es 86
Fi gur e 3.13 oper at i ng pr of i l e of a gener i c i t em
The sci ence of f ai l ur es has not advanced suf f i ci en t l y, as yet , t o b e abl e t o
pr edi ct f ai l ur e t i me di st r i but i on i n al l cases. Thi s i s cur r ent l y done
empi r i cal l y by r unni ng a sampl e of i t ems on t est unt i l t hey f ai l , or f or an
ext ended p er i od, usual l y under ‘ i deal ’ condi t i ons t hat at t empt t o si mul at e
t he o per at i onal envi r onment . M i l i t ar y ai r cr af t  engi nes, f or exampl e, ar e
exp ect ed t o oper at e whi l e subj ect ed t o f or ces bet ween  5 and + 9 ‘ g’ ,
al t i t udes f r om zer o t o 50000 f eet (15000 met er s) and speeds f r om zer o t o
M ach 2+. One has t o t est t he equi pment w i t h some new and some ol d
compon en t s t o f i nd t he exp ect ed val ues of t he r ando m var i abl es X
2
, X
3
, et c.
In pr act i ce most of t h e t est i ng i s done on new i t ems wi t h al l new
compon en t s i n pr i st i ne condi t i on. The val ue der i ved by t h ese t yp e of
t est i ng wi l l gi ve t he expect ed val ue of t he r andom var i abl e X
1
. In pr act ice,
t he expect ed val ue of X
1
i s quot ed as M TBF. I n f act , t he exp ect ed val ue of X
1
w i ll gi ve onl y t he M ean Ti me To Fi r st Fai l ur e (as t he t est i ng i s done on new
i t ems and t he t i mes r ef l ect t h e t i me t o f i r st f ai l ur e) and not t h e M TBF. To
cal cul at e M TBF one shoul d consi der t he expect ed val ues of t he r andom
var i abl es X
2
, X
3
, et c.
If t he t i me t o f ai l ur e di st r i but i on of t he syst em i s exponent i al t hen t he
M TBF can be est i mat ed usi ng t he f ol l owi ng equat i on (r ecommend ed by
M ILHDBK217 and DEFSTAN0040):
MTBF
T
n
· (3.24)
wher e, T i s t he t ot al op er at i n g p er i od and `n’ i s t he number of f ai l ur es
dur i ng t hi s per i od. Not e t hat t he above r el at i on i s val i d onl y f or l ar ge val ue
of T. If n = 0, t hen M TBF becomes i nf i ni t y, t hus one shoul d b e car ef ul i n
usi ng t he above r el at i on. The above expr essi on can be used onl y when
suf f i ci ent amount of dat a i s avai l abl e.
Oper at i n g X
1
Down
t i me
Dow n
t i me
Oper at i ng X
1
3. Rel i abi l i t y M easur es 87
Charact eri st i cs of MTBF
1. The value of MTBF is equal to MTTF if after each repair the system is as
good as new.
2. MTBF = 1 / λ for exponential distribution, where λ is the scale parameter
( also the hazard function ).
Appl i cat i ons of MTBF
1. For a repairable system, MTBF is the average time in service between
failures. Note that, this does not include the time spent at repair facility
by the system.
2. MTBF is used to predict steadystate availability measures like inherent
and operational availability.
3.18. PERCENTILE LIFE ( TTF
P
OR B
P%
)
Per cent i l e l i f e or B
p %
i s a measur e of r el i abi l i t y whi ch i s popul ar among
i ndust r i es. Thi s i s t he l i f e by whi ch cer t ai n pr opor t i on of t he popul at i on (p
%) can be expect ed t o have f ai l ed. B
10%
means t he l i f e (t i me) by whi ch 10%
of t he i t ems wi l l be expect ed t o have f ai l ed. Per cent i l e l i f e i s now
f r equ ent l y used amo ng aer ospace i ndust r i es as a desi gn r equi r ement .
M at hemat i cal l y p er cent i l e l i f e can be obt ai ned by sol vi ng t he f ol l owi ng
equat i on f or t :
F t f x dx p
t
( ) ( ) % · ·
∫
0
(3.25)
Assume t hat F(t ) i s a exponent i al di st r i but i on wi t h paramet er λ = 0.05, and
we ar e i nt er est ed i n f i ndi ng B
10
. Then f r om above equat i on w e have:
1 005 010 2107 − − · ⇒ · exp( . ) . . t t
Thus 2. 107 i s t he B
10
l i f e f or exponent i al di st r i but i on wi t h par amet er 0.05.
The mai n appl i cat i on of per cent i l e l i f e li es i n pr edi ct i on of i ni t i al spar es
r equi r ement (i ni t ial spar es pr ovi si oni ng, IP).
3. Rel i abi l i t y M easur es 88
Sy st em Rel i ab i l i t y
an d M ai n t ai n ab i l i t y
Co u r se m at er i al
Cour se I nst r uct or : Pr of essor U Di nesh Kumar
Indi an Inst i t ut e of M anagement Bangal or e
4. Syst ems Rel i abi l i t y 89
Chapter 4
Systems Reliability
' A Bi r d i s an i nst r ument wor ki ng accor di ng t o a mat hemat i cal l aw . It l i es wi t hi n
t he power of man t o make t hi s i nst r ument wi t h al l i t s mot i on'
Leonar do da Vi nci
In t hi s chapt er , we pr esent met hodol ogi es t hat can be used t o eval uat e
syst ems r el i abi l i t y usi ng si mpl e mat h emat i cal t ool s. The chapt er di scusses
t wo appr oach es t hat can be used t o pr edi ct t he r el i abi l i t y met r i cs of t he
syst em. Fi r st , we st udy t he model s t hat ar e based on si mpl e pr obabi l i t y
t heor y, assumi ng t hat t he t i met of ai l ur e di st r i but i ons of di f f er ent
component s w i t hi n t he syst em ar e known. These model s can be used onl y
f or nonr epai r abl e i t ems. The second appr oach i s based on M ar kov model s,
f or pr edi ct i ng di f f er ent r el i abi l i t y measu r es. The model s f or r epai r abl e
i t ems wi l l be di scussed usi ng t he M ar kov model s. Thr oughout t he Chapt er ,
t he wor d ‘ syst em’ i s used t o r epr esent t he compl et e equi pmen t and t he
wor d ‘ i t em’ i s used as a gen er i c t er m t hat st ands f or subsyst em, modul e,
component , par t or uni t . Any r el i abil i t y pr edi ct i on met hodol ogy usi ng t i me
t of ai l ur e appr oach wi ll i nvol ve t he f ol l owi ng st eps:
1. Const r uct t he r el i abi l i t y bl o ck di agr am (RBD) of t he syst em. Thi s may
i nvol ve per f or mi ng f ai l ur e modes and ef f ect anal ysi s (FM EA).
2. Det er mi ne t he oper at i onal pr of i l e of each bl ock i n t he r el i abi l i t y bl ock
di agr am.
3. Der i ve t he t i met of ai l ur e di st r i but i on of each bl ock.
4. Der i ve t h e l i f e exchange r at e mat r i x (LERM ) f or t he di f f er en t
component s w i t hi n t he syst em.
5. Comput e r el i abi l i t y f unct i on of each bl ock.
4. Syst ems Rel i abi l i t y 90
6. Comput e t he r el i abi l i t y f unct i on of t he syst em.
4.19. RELIABILITY BLOCK DIAGRAM
Rel i abi l i t y bl o ck di agr am, RBD, of an i t em i s a l ogi cal di agr ammat i c
i l l ust r at i on of t he syst em i n w hi ch each i t em (har dw ar e/ sof t w ar e) w i t hi n
t he syst em i s r epr esen t ed b y a bl o ck. RBD f or ms a basi s f or cal cul at i on of
syst em r el i abi l i t y measur es. Each bl ock wi t hi n a RBD can r epr esen t a
componen t , subsyst em, modul e or syst em. The st r uct ur e of a RBD i s
det er mi ned b y t he ef f ect of f ai l ur e of each bl ock on t he f unct i onal i t y of t he
syst em as a w hol e. A bl ock does not have t o r epr esent physi cal l y
connect ed har dw ar e i n t he act ual syst em t o b e connect ed i n t he bl ock
di agr am. I n an RBD t he i t ems whose f ai l ur e can cause syst em f ai l ur e
i r r espect i ve of t he r emai ni ng i t ems of t he syst em ar e connect ed i n ser i es.
It ems whose f ai l ur e al one cannot cause syst em f ai l ur e ar e connect ed i n
par al l el . Dependi n g on t he i t em, a RBD can be r epr esent ed by a ser i es,
par al l el , ser i espar al l el , r out of n or compl ex conf i gur at i on. Const r uct i on of
RBD r equi r es f unct i onal anal ysi s of var i ous par t s wi t hi n t he syst em. Each
bl ock wi t hi n a RBD shoul d be descr i bed usi ng t i met of ai l ur e di st r i but i on f or
t he pur pose of cal cul at i ng syst em r el i abi l i t y measur es. Th e RBD can al so
have net w or k st r uct ur es (e.g. communi cat i on syst ems, w at er n et w or k and
Int er net ). In t he f ol l owi ng sect i ons we addr ess how t o eval uat e var i ous
r el i abi l i t y measur es f or di f f er ent r el i abi l i t y bl ock di agr ams.
4.20. RELIABILITY MEASURES FOR SERIES
CONFIGURATION
In a ser i es conf i gur at i on, al l t he consi st i ng i t ems of t he syst em sh oul d be
avai l abl e or f unct i onal t o mai nt ai n t he r equi r ed f unct i on of t he syst em.
Thus, f ai l ur e of any one i t em of t h e syst em w i l l cause f ai l ur e of t he syst em
as whol e. Ser i es conf i gur at i on i s pr obabl y t he most commonl y encount er ed
RBD i n engi neer i ng pr act i ce. The RBD of a hypot het i cal syst em whose i t ems
ar e connect ed i n ser i es i s gi ven i n Fi gur e 4.1.
Fi gur e 4.1. Rel i abi l i t y bl ock di agr am of a syst em wi t h ser i es conf i gur at i on
Rel i abi l i t y f unct i on of seri es conf i gurat i on
1 2 n
n
…
4. Syst ems Rel i abi l i t y 91
Rel i abi l i t y f unct i on of a syst em wi t h ser i es conf i gur at i on can b e der i ved
f r om t he r el i abi l i t y f unct i on of i t s consi st i ng i t ems. Let R
S
(t ) r epr esent t he
r el i abi l i t y f unct i on of a ser i es syst em wi t h n i t ems. Let R
i
(t ) denot e t he
r el i abi l i t y f unct i on of t h e i t em i . If TTF
i
i s t he t i me t of ai l ur e r andom
var i abl e f or t he i t em i , t hen t he r el i abi l i t y f unct i on of syst em f or ‘ t ’ hour s of
oper at i on i s gi ven by:
R
S
(t) = P [ TTF
1
≥ t, TTF
2
≥ t, ...., TTF
n
≥ t ] (4.1)
The equat i on (4.1) cl ear l y st at es t hat t he syst em under consi der at i on wi l l
mai nt ai n t he r equi r ed f unct i o n i f and onl y i f al l t he n i t ems of t h e syst em
ar e abl e t o mai nt ai n t he r equi r ed f un ct i on f or at l east t hour s of op er at i on.
Assumi ng t hat t he r andom var i abl es TTF
i
ar e i ndependent of each ot her ,
t he expr essi on (4.1) can be w ri t t en as:
R
s
(t ) = P[ TTF
1
≥ t ] × P[ TTF
2
≥ t ] × ... × P[ TTF
n
≥ t ]
= R
1
(t ) × R
2
(t ) × .. . × R
n
(t )
Thus, t he r el i ab i l i t y of a ser i es conf i gur at i on wi t h n i t ems i s gi ven by:
R t R t
s i
i
n
( ) ( ) · ∏
·1
(4.2)
Not e t hat i n t he ab ove equat i on (4.2), i t i s assu med t hat t h e connect i ng
medi a (such as sol der j oi nt s) bet w een di f f er ent i t ems i s 100% r el i abl e
(unl ess t hi s i s speci f i cal l y i ncl uded i n t he RBD). However , t hi s need not be
t r ue. In t he equat i on (4.2) t i me t i s used as a gener i c t er m. I n most case
t ime act ual l y r epr esent s age or ut i l i sat i on of t he i t em under consi d er at i on.
It can have di f f er ent uni t s such as hour s, mi l es, l andi ngs, cycl es et c f or
di f f er en t i t ems. One has t o nor mal i se t he ‘ t i me’ bef or e cal cul at i ng t he
r el i abi l i t y f unct i on i n such cases. One met h od of nor mal i si ng t he di f f er en t
l i f e uni t s of t he i t ems i s usi ng Li f e Exchange Rat e M at r i x (LERM ), whi ch wi l l
be di scussed l at er i n t hi s chapt er . When t he l i f e uni t s of i t ems ar e di f f er ent
(or di f f er ent i t ems have di f f er ent ut i l i sat i on), w e use t he f ol l ow i ng equat i on
t o f i nd t he r el i abi l i t y of t he ser i es syst em.
R P TTF t TTF t TTF t R t R t R t
s n n n n
· ≥ ≥ ≥ · × × × [ , , , ] ( ) ( ) ( )
1 1 2 2 1 1 2 2
L L
That i s,
4. Syst ems Rel i abi l i t y 92
∏
·
·
n
i
i i S
t R t R
1
) ( ) ( (4.3)
In equat i on (4.3), t
i
i s t he age of t h e i t em i , w hi ch i s equi val ent t o age t of
t he syst em. That i s, f or t he syst em t o sur vive up t o age t , t he i t em i shoul d
sur vi ve up t o t
i
. Thr ough out t hi s book we use equat i on (4.3) unl ess
ot her wi se sp eci f i ed.
Charact eri st i cs of rel i abi l i t y f unct i on of a seri es conf i gur at i on
1. The val u e of t h e r el i abi l i t y f unct i on of t he syst em, R
S
(t ), f or a ser i es
conf i gur at i on i s l ess t han or equal t o t he mi ni mum val ue of t he i ndi vi dual
r el i abi l i t y f unct i on of t he const i t ut i ng i t ems. That i s:
R t Min R t
S
i n
i
( ) { ( )}
,2,..
≤
·1
2. If h
i
(t ) r epr esent t he hazar d f unct i on of i t em i , t h en t h e syst em r el i abi l it y
of a ser i es syst em can be wr i t t en as:
dx x h
dx x h t R
t n
i
i
n
i
t
i s
] ) ( [ exp(
) ( exp( ) (
0 1
1 0
∫ ∑ − ·
∏ ∫
− ·
·
·
Exampl e 4.1
A syst em consi st s of f our i t ems, each of t hem ar e necessar y t o mai nt ai n t he
r equi r ed f unct i on of t h e syst em. The t i me t o f ai l ur e di st r i but i on and t hei r
cor r espondi ng par amet er val u es ar e gi ven i n Tabl e 4.1. Fi nd t he r el i abi l i t y
of t he syst em f or 500 and 750 hour s of oper at i on.
Tabl e 4.1 Ti me t o f ai l ur e di st ri but i on and t hei r par amet er of t he i t ems
It em Ti me t o f ai l ur e di st r i but i on Par amet er val ues
It em 1 Exponent i al λ = 0.001
It em 2 Wei bul l η = 1200 hour s β = 3. 2
It em 3 Nor mal µ = 800 hour s σ = 350
It em 4 Wei bul l η = 2000 hour s β = 1. 75
4. Syst ems Rel i abi l i t y 93
SOLUTION:
Fr om t h e i nf or mat i on gi ven i n Tabl e 4.1, t he r el i abi l i t y f unct i on of var i ous
i t ems can be w r i t t en as:
R t t
R t
t
R t
t
R t
t
1
2
3 2
3
4
175
0 001
1200
800
350
2000
( ) exp( . )
( ) exp[ ( ) ]
( ) ( )
( ) exp[ ( ) ]
.
.
· − ×
· −
·
−
· −
Φ
Si nce t he i t ems ar e connect ed i n ser i es, t he r el i abi l i t y f unct i on of t he
syst em i s gi ven by:
R t t
t t t
s
( ) exp( . ) exp[ ( ) ] ( ) exp[ ( ) ]
. .
· − × × − ×
−
× − 0 001
1200
800
350 2000
3 2 175
Φ
Fi gur e 4.2 Rel i abi l i t y f unct i on of t he syst em and i t s const i t uent i t ems.
Subst i t ut i ng t = 500 and 750 i n t he above equat i on, we get :
R(500) = 0.6065 × 0.9410 × 8043 × 0.9154 = 0.4202
R(750) = 0.4723 × 0.8003 × 0.5568 × 0. 8355 = 0.1759
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
2
0
0
4
0
0
6
0
0
8
0
0
1
0
0
0
1
2
0
0
1
4
0
0
1
6
0
0
1
8
0
0
2
0
0
0
2
2
0
0
2
4
0
0
2
6
0
0
2
8
0
0
3
0
0
0
3
2
0
0
3
4
0
0
3
6
0
0
3
8
0
0
4
0
0
0
Time
R
e
l
i
a
b
i
l
i
t
y
f
u
n
c
t
i
o
n
R
4
(t )
R
3
(t )
R
1
(t )
R
2
(t )
R
s
(t )
4. Syst ems Rel i abi l i t y 94
Fi gur e 4.2 shows t he r el i abi l i t y f unct i on of t he syst em and var i ous i t ems of
t he syst em. Not e t hat t h e syst em r el i ab i l i t y val u e i s al ways l ess t han or
equal t o any of t he const i t ut i ng i t ems.
Exampl e 4.2
Avi oni cs syst em of an ai r cr af t consi st s of di gi t al aut op i l ot , i nt egr at ed gl obal
posi t i oni ng syst em, weat her and gr ound mappi ng r adar , di gi t al map di spl ay
and war ni ng syst em. Apar t f r om t he above i t ems, t he avi oni cs syst em has
cont r ol sof t war e. The t i met of ai l ur e di st r i but i ons of var i o us i t ems ar e gi ven
i n Tabl e 4.2. Fi nd t he r el i abi l i t y of t he avi oni cs syst em f or 100 hour s of
oper at i on i f al l t he i t ems ar e n ecessar y t o mai nt ai n t he r equi r ed f unct i on of
t he avi oni cs syst em.
Tabl e 4.2 Ti met of ai l ur e di st r i but i on of var i ous i t ems of t he avi oni cs syst em
It em Ti met of ai l ur e
di st r i but i on
Par amet er val ues
Di gi t al aut opi l ot Exponent i al λ = 0.003
Int egr at ed gl obal
posi t i oni ng syst em
Wei bul l η = 1200, β = 3.2
Weat her and gr ound
mappi ng r adar
Wei bul l η = 1000, β = 2.1
Di gi t al map di spl ay Nor mal µ = 800, σ = 120
War ni ng Syst em Nor mal µ = 1500, σ = 200
Sof t war e Exponent i al λ = 0.001
SOLUTION:
Fr om t he dat a gi ven i n Tabl e 4.2, we can d er i ve t h e r el i abi l i t y f unct i on of
var i ous i t ems as f ol l ows:
1. Rel i abi l i t y of di gi t al aut opi l o t
R t t R
1 1
100 0 003 100 07408 ( ) exp( ) ( ) exp( . ) . · − × ⇒ · − × · λ
4. Syst ems Rel i abi l i t y 95
2. Rel i abi l i t y of i nt egr at ed gl obal posi t i oni ng syst em.
R t R
2 2
3 2
100 100 100 1200 09996 ( ) exp( ( / ) ) ( ) exp( ( / ) ) .
.
· − ⇒ · − · η
β
3. Rel i abi l i t y of weat her and gr ound mappi ng syst em r adar
R t R
3 3
2 1
100 100 100 1000 0 9920 ( ) exp( ( / ) ) ( ) exp( ( / ) ) .
.
· − ⇒ · − · η
β
4. Rel i abi l i t y of di gi t al map di spl ay
R
t
R
4 4
100 100
800 100
120
58 1 ( ) ( ) ( ) ( ) ( . ) ·
−
⇒ ·
−
· · Φ Φ Φ
µ
σ
5. Rel i abi l i t y of war ni ng syst em
R
t
R
5 4
100 100
1500 100
200
7 1 ( ) ( ) ( ) ( ) ( ) ·
−
⇒ ·
−
· · Φ Φ Φ
µ
σ
6. Rel i abi l i t y of sof t war e
R t t
6
0001 100 0 9048 ( ) exp( ) exp( . ) . · − ⇒ − × · λ
Thus, t he r el i abi l i t y of t he avi oni cs syst em f or 100 hour s of op er at i on i s
gi ven by:
R R
s i
i
( ) ( ) . . . . . 100 100 0 7408 09996 09920 1 1 09048 06646
1
6
· · × × × × × ·
·
∏
Hazard f unct i on of a seri es conf i gur at i on
Let R
S
(t ) denot e t h e r el i abi l i t y f unct i on of t he syst em. Fr om def i ni t i on, t he
hazar d r at e of t he syst em, h
S
(t ), can be wr i t t en as:
h t
dR t
dt R t
S
S
S
( )
( )
( )
· − ×
1
(4.4)
Usi ng equat i on (4.2), t he expr essi on f or R
S
(t ) can be wr i t t en as:
4. Syst ems Rel i abi l i t y 96
R t R t F t
S i
i
n
i
i
n
( ) ( ) [ ( )] · · −
· ·
∏ ∏
1 1
1 (4.5)
wher e F
i
(t ) i s t he f ai l ur e f unct i on of t he i t em i . Di f f er ent i at i ng t h e above
expr essi on f or r el i abi l i t y f unct i on wi t h r espect t o t , w e get :
dR t
dt
f t F t
i
i
n
i
j
j i
n
( )
( ) [ ( )] · − −
· ·
≠
∑ ∏
1 1
1 (4.6)
Subst i t ut i ng equat i on (4.6) i n equat i on (4.4), we get
h t
f t
R t
h t
S
i
i
i
n
i
i
n
( )
( )
( )
( ) · ·
· ·
∑ ∑
1 1
(4.7)
4. Syst ems Rel i abi l i t y 97
Tabl e 4.3 Hazar d r at e of ser i es conf i gur at i on wi t h n i t ems.
Pr obabi l i t y densi t y f unct i on
of i t h i t em, f
i
(t )
Hazar d f unct i on of t he syst em, h
S
(t )
(Exponen t i al )
) exp( t
i i
λ λ −
∑
·
·
n
i
i S
t h
1
) ( λ
(Weibull)
) ) ( exp( ) (
1
i i
i i i
i
t t
β β
η η η
β
−
−
∑
·
−
·
n
i i i
i
S
i
t
t h
1
1
) )( ( ) (
β
η η
β
(Normal)
) ) (
2
1
( exp(
2
1
2
i
i
i
t
σ
µ
π σ
−
−
∑
·
−
Φ ·
n
i i
i
i S
t
t f t h
1
) ( ) ( ) (
σ
µ
Thus t he hazar d f unct i on of a ser i es syst em i s gi ven by t he sum of t he
hazar d f unct i on of i ndi vi dual i t ems. Tabl e 4.3 gi ves hazar d f unct i on of a
ser i es conf i gur at i on wi t h n i t em und er t he assu mpt i on t hat t he t i met o
f ai l ur e of t he i t ems f ol l o ws same di st r i but i on but have di f f er en t par amet er .
Fi gur e 4.3 show s hazar d r at e of a ser i es syst em w i t h t w o i t ems w her e t he
t i met of ai l ur e of i ndi vi dual i t ems f ol l ow Wei bul l di st r i but i on.
0
1
2
3
4
5
6
7
0
0
.
2
0
.
4
0
.
6
0
.
8 1
1
.
2
1
.
4
1
.
6
1
.
8 2
2
.
2
2
.
4
Time
h
a
z
a
r
d
r
a
t
e
h
s
(t )
η = 1, β = 2.1
η = 1, β = 0.5
4. Syst ems Rel i abi l i t y 98
Fi gur e 4. 3 Hazar d r at e of ser i es syst em wi t h t wo i t ems wi t h Wei bul l t i me
t of ai l ur e di st r i but i on.
In most cases, t h e hazar d f unct i on of a ser i es conf i gur at i on w i l l be a
i ncr easi ng f unct i on. For exampl e, consi der a ser i es syst em wi t h 10 i t ems.
Let 9 out of 10 i t ems be i dent i cal and have exponent i al t i met of ai l ur e
di st r i but i on wi t h par amet er wi t h r at e λ = 0.01. Now w e consi der t w o
di f f er en t cases f or t h e t i met of ai l ur e di st r i but i on of t he r emai ni ng one
i t em.
Fi gur e 4.4 Hazar d r at e t he syst em w i t h 10 i t ems w her e 9 of t hem have
const ant hazar d.
Case 1:
Let t he t i met of ai l ur e of t he r emai ni ng one i t em be r epr esent ed by usi ng
Wei bul l di st r i but i on w i t h scal e par amet er η = 100 and β = 2. 5. Now t he
hazar d r at e of t hi s syst em i s gi ven by:
h t
t
s
( ) . ( ) · × +
−
9 0 01
1
β
η η
β
It i s obvi ous f r om t he above expr essi on t hat t he hazar d r at e of t he syst em i s
not const ant . Fi gur e 4.4 shows t he ef f ect of nonconst ant hazar d f unct i on
on t he syst em hazar d f unct i on even when most of t h e i t ems have const ant
hazar d f unct i on. In Fi gur e 4.4, h
1
(t ) r epr esent s t he hazar d r at e f or t he ni ne
i t ems wi t h exponent i al t i met of ai l ur e and h
2
(t ) r epr esent t he hazar d r at e
of t he i t em w i t h Wei bul l t i met of ai l ur e di st r i but i on.
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0
1
5
3
0
4
5
6
0
7
5
9
0
1
0
5
1
2
0
1
3
5
1
5
0
1
6
5
1
8
0
1
9
5
2
1
0
2
2
5
2
4
0
Time
H
a
z
a
r
d
r
a
t
e
h
1
(t)
h
2
(t)
h
s
(t)
4. Syst ems Rel i abi l i t y 99
Let t he t i me t of ai l ur e of t he r emai ni ng one i t em can be r epr esent ed by
usi ng Wei bul l di st r i but i on wi t h scal e par amet er η = 100 and β = 0. 5. Now
t he hazar d r at e of t hi s syst em i s gi ven by:
h t
t
s
( ) . ( ) · × +
−
9 0 01
1
β
η η
β
It i s obvi ous f r om t he above expr essi on t hat t he hazar d r at e of t he syst em i s
not const ant . Fi gur e 4.5 shows t he ef f ect of nonconst ant hazar d f unct i on
on t he syst em hazar d f unct i on even when most of t h e i t ems have const ant
hazar d f unct i on. In Fi gur e 4.5, h
1
(t ) r epr esent t he hazar d r at e f or t he ni ne
i t em w i t h exponent i al t i met of ai l ur e and h
2
(t ) r epr esent t he hazar d r at e of
t he i t ems w i t h Wei bul l t i met of ai l ur e di st r i but i on.
Not e: The hazar d f unct i on of compl ex r epai r abl e syst em may conver ge t o a
const ant hazar d f unct i on under cer t ai n condi t i ons (mai nl y under st eady
st at e condi t i ons). Thi s r esul t pr oved by Dr eni ck (1961) may not be t r ue f or
t oday’ s hi ghl y r el i abl e syst ems. Thus, one has t o be ver y car ef ul i n usi ng
const ant hazar d f unct i on and t hus exponent i al t i me t o f ai l ur e f or compl ex
syst ems. Thi s pr obl em wi l l be f ur t her di scussed i n Chapt er 8.
0
0. 02
0. 04
0. 06
0. 08
0. 1
0. 12
0
1
5
3
0
4
5
6
0
7
5
9
0
1
0
5
1
2
0
1
3
5
1
5
0
1
6
5
1
8
0
1
9
5
2
1
0
2
2
5
2
4
0
Time
h
a
z
a
r
d
f
u
n
c
t
i
o
n
h
s
(t)
h
1
(t)
h
2
(t)
4. Syst ems Rel i abi l i t y 100
Fi gur e 4.5 Hazar d f unct i on of t he syst em wi t h 10 i t ems wher e 9 of t hem
have const ant hazar d.
Exampl e 4.3
A syst em has t w o i t ems A and B connect ed i n ser i es. The t i met of ai l ur e of
i t em A f ol l ow s exponent i al di st r i but i on w i t h par amet er λ = 0.002. Th e t i me
t of ai l ur e of i t em B f ol l ow s Wei bul l di st r i but i on wi t h par amet er η = 760 and
β = 1. 7. Fi nd t he hazar d r at e of t hi s syst em at t i me t = 100 and t = 500.
SOLUTION:
Let h
A
(t ) and h
B
(t ) r epr esent t h e hazar d r at e of i t em A and B r espect i vel y.
Si nce t h e i t ems ar e co nnect ed i n ser i es, t he hazar d r at e of t he syst em, h
S
(t )
i s gi ven by:
7 . 0 1
)
760
)(
760
7 . 1
( 002 . 0 ) )( ( ) ( ) ( ) (
t t
t h t h t h
B A S
+ · + · + ·
− β
η η
β
λ
Subst i t ut i ng t = 100 and t = 500 i n t he above equat i on,
h
S
(100) = 0.00254
h
S
(500) = 0.0036
Mean t i me t o f ai l ure of a seri es conf i gurat i on
The mean t i me t o f ai l ur e, M TTF, of a ser i es conf i gur at i on, denot ed by
M TTF
S
, can be wr i t t en as:
∫ ∏ ·
∫
·
∞
·
∞
0 1 0
) ( dt t R dt R MTTF
n
i
i S S
(4.8)
The above i nt egr al can be eval uat ed usi ng numer i cal i nt egr at i on i f t he
f ai l ur e di st r i but i on i s Wei bul l , nor mal , l ognor mal or Gamma. How ever , i n
case of exponen t i al di st r i but i on t he expr essi on f or syst em M TTF
S
can be
obt ai ned as f ol l ow s. Assume t hat t he t i met of ai l ur e di st r i but i on of
component i i s gi ven by, 1− − exp( ). λ
i
t Subst i t ut i ng R t t
i i
( ) exp( ) · −λ i n
equat i on (4.8) w e have,
4. Syst ems Rel i abi l i t y 101
∫ ∫ ∑ ∏ ∫ ∏
∞ ∞
· ·
∞
·
− · − · ·
0 0 1 1 0 1
) exp( ) exp( ) ( dt t dt t dt t R MTTF
n
i
i
n
i
i
n
i
i s
λ λ
∑
·
·
n
i
i
s
MTTF
1
1
λ
(4.9)
Thus, t he M TTF
S
of a ser i es conf i gur at i on wi t h n i t ems wher e t he t i met o
f ai l ur e of t he i t ems ar e r epr esent ed by exponent i al di st r i but i on i s gi ven by
t he i nver se of t he syst em’ s hazar d f unct i on. Not e t hat t hi s r esul t i s t r ue
onl y when t he t i met of ail ure di st ri but i on i s exponent i al . The f ol l ow i ng
equat i on der i ved usi ng t r apezi um appr oxi mat i on of equat i on (4.8) can be
used whenever t h e t i met of ai l ur e of at l east one i t em i s nonexponent i al .
] [ ]) * [ ] 0 [ (
2
1
1
h i R h h M R R
h
MTTF
M
i
S
× × + + × ≈
∑
−
·
(4.10)
Wh er e h i s a smal l val ue (e.g. 0.01 or 0.1), t he val ue of M i s sel ect ed such
t hat ) ( h M R
S
× i s al most zer o.
Exampl e 4.4
A syst em consi st s of t hr ee i t ems connect ed i n ser i es. The t i met of ai l ur e
di st r i but i on and t hei r cor r esp ondi ng par am et er val ues ar e gi ven i n Tabl e
4.4. Fi nd t he mean t i me t o f ai l ur e of t he syst em. Compar e t he val ue of
M TTF
S
wi t h mean t i me t o f ai l ur e of i ndi vi dual i t ems.
Tabl e 4.4 Ti met of ai l ur e di st r i but i on of di f f er ent i t ems
I t em Di st r i but i on Par amet er val ues
I t em 1 Wei bul l η
1
= 10, β
1
= 2. 5
I t em 2 Exponen t ial λ = 0.2
I t em 3 Wei bul l η
2
= 20, β
2
= 3
4. Syst ems Rel i abi l i t y 102
SOLUTION:
M ean t i me t o f ai l ur e of t he syst em i s gi ven by:
dt
t
t
t
dt t R MTTF
i
i S
) ) ( exp( ) exp( ) ( exp(
) (
2 1
2 1 0
0
3
1
β β
η
λ
η
− × − × − ·
·
∫
∫ ∏
∞
∞
·
∫
∞
− × − × − ·
0
3 5 . 2
) )
20
( exp( ) 2 . 0 exp( ) )
10
( exp( dt
t
t
t
MTTF
S
Usi ng numer i cal i nt egr at i on, t he M TTF
S
i s gi ven by:
M TTF
S
≈ 3.48
Tabl e 4. 5 gi ves t h e mean t i me t o f ai l ur e of var i ous i t ems. Not e t hat t he
mean t i me t o f ai l ur e of t he syst em i s al ways l ess t han t hat of t he
compon en t s when t he i t ems ar e connect ed i n ser i es.
Tabl e 4.5 Compar i son of M TTF of i ndi vi dual i t ems and M TTF
S
I t em 1 I t em 2 It em 3 Syst em
M TTF = 8. 87 M TTF = 5 M TTF = 17.86 M TTF
S
≈ 3.48
Char act er i st i cs of MTTF
S
of seri es syst em
1. The M TTF
S
≤ M TTF
i
, w her e M TTF
i
i s t he mean t i me t o f ai l ur e of t he i t em
i . Thus, t he mean t i m e t o f ai l ur e of a syst em wi t h ser i es RBD wi l l be l ess
t han t he mean t i me t o t i me f ai l ur e of any of i t s const i t ut i ng i t ems.
MTTF Min MTTF
S
i n
i
≤
·1,2,...,
{ }
Wh er e M TTF
i
d enot e t he mean t i me t o f ai l ur e of t he i t em i .
4. Syst ems Rel i abi l i t y 103
2. For compl ex r epai r abl e syst ems, M TTF
S
, r epr esent s t he mean t i me t o f i r st
f ai l ur e.
4.21. LIFE EXCHANGE RATE MATRIX
Not al l t he co mponent s of t he i t em wi l l have t he same ut i l i sat i on or l i f e
uni t . In some cases, i f t he act ual mi ssi on per i od i s t hour s, some i t ems of
t he syst em may have t o oper at e mor e t han t hour s (i n many cases i t can be
l ess t han t ho ur s). An ai r cr af t j et en gi ne wi l l be swi t ch ed on at l east 20
mi nut es b ef or e t he act ual f l i ght . Thus, f or 10 hour s f l i ght , t he engi n e may
have t o oper at e f or mor e t han 10 hour s. Oper at i onal envi r onment can al so
change t he agei ng pat t er n of di f f er en t compon ent s w i t hi n a syst em. For
exampl e, t he aver age f l i ght of a domest i c f l i ght wi t hi n Japan i s ar ound 30
mi nut es compar ed t o t hat of ar ound 3 hour s i n US. Thus t he ai r cr af t used i n
Japan l ands mor e of t en t han t he on e i n USA. Thi s means t hat t he usage of
l andi ng gear s, t yr es et c of ai r cr af t used i n domest i c f l i ght s i n Japan wi l l be
much hi gher t han t hat of USA. It i s ver y common t hat di f f er ent i t ems w i t hi n
a syst em may have di f f er ent l i f e uni t s such as hour , mi l es, f l yi ng hour s,
l andi ngs, cycl es et c. Thus, t o f i nd t h e r el i abi l i t y of a syst em w hose i t ems
have di f f er ent l i f e uni t s i t i s n ecessar y t o nor mal i se t he l i f e uni t s. In t hi s
sect i on we i nt r oduce t he concept of l i f e exchange r at e mat r i x, whi ch can be
used t o descr i b e t he exchange r at es bet w een var i ous l i f e uni t s.
Li f e exchange r at e mat r i x (LERM ) i s a squ ar e mat r i x of si ze n, w her e n i s t he
number of i t ems i n t he syst em. Let us denot e t h e l i f e exchange r at e mat r ix
as R = [ r
i, j
] , w her e r
i ,j
i s t he (i ,j ) t h el ement i n t he LERM . Thus, f or a syst em
wi t h n i t ems connect ed i n ser i es, t he LERM can be r epr esent ed as:
1
1
1
1
1
1
1
]
1
¸
·
n n n n
n
n
r r r
r r r
r r r
LERM
, 2 , 1 ,
, 2 2 , 2 1 , 2
, 1 2 , 1 1 , 1
...
.
.
.
.
.
.
.
.
.
.
.
.
...
...
The el emen t s of LERM ar e i nt er pr et ed as f ol l ows:
r
i ,j
denot es t hat :
4. Syst ems Rel i abi l i t y 104
1 li f e uni t of i = r
i,j
× 1 l i f e uni t of j .
Any LERM wi l l sat i sf y t he f ol l owi ng condi t i ons:
1
,
·
i i
r for all i.
j k k i j i
r r r
, , ,
× · for all i, j, k
i j
j i
r
r
,
,
1
·
As an exampl e, l et us consi der a syst em wi t h t hr ee i t ems connect ed i n
ser i es (Fi gur e 4.6). Let t he l i f e uni t of i t ems 1, 2 and 3 be hour s, mi l es and
cycl es r esp ect i vel y.
Fi gur e 4.6. Ser i es syst em wi t h t hr ee i t ems wher e each i t em has di f f er en t l i f e
uni t s
Assume t hat :
1 hour = 10 mi l es
1 hour = 5 cycl es
Usi ng t he abo ve dat a, i t i s easy t o const r uct t h e l i f e exchange r at e mat r i x
f or t he above syst em. The LERM f or t he above mat r i x i s:
1
1
1
]
1
¸
·
1 2 5 / 1
5 . 0 1 10 / 1
5 10 1
R
One can easi l y ver i f y t hat t h e above mat r i x sat i sf i es al l t hr ee condi t i ons f or
a l i f e exchange r at e mat r i x. Usi ng t he above mat r i x, one can easi l y measur e
1
Hours
2
Miles
3
Cycles
4. Syst ems Rel i abi l i t y 105
r el i abi l i t y char act er i st i cs i n nor mal i sed l i f e uni t . For t he RBD shown i n
Fi gur e 4.6, r el i abi l i t y of t he syst em f or 5 cycl es i s gi ven by R
1
(1)×R
2
(1)×R
3
(5).
Example 4.5
Rel i abi l i t y bl ock di agr am of a syst em consi st s of t hr ee modul es A, B and C
connect ed i n ser i es. The t i me t of ai l ur e of modul e A f ol l ows Wei bul l
di st r i but i on wi t h scal e par amet er η = 100 hour s and β = 3.2. The t i met o
f ai l ur e of modul e B f ol l ows Nor mal di st ri but i on wi t h par amet er µ = 400
cycl es and σ = 32 cycl es. The t i m et of ai l ur e o f modul e C f ol l ow s
expon ent i al di st r i but i on wi t h par amet er λ = 0.00015 per mi l e. It was al so
not ed t hat , dur i ng 1 hour , t he modul e B per f or ms 12 cycl es and modul e C
per f or ms 72 mi l es. Fi nd t he pr obabi l i t y t hat t he syst em wi l l sur vi ve up t o
240 cycl es of modul e B.
SOLUTION:
For t h e syst em t o sur vi ve 240 cycl es, modul e A sho ul d sur vi ve up t o 20
hour s and modul e C shoul d sur vi ve up t o 1440 mi l es.
The r el i abi l i t y of i ndi vi dual modul es ar e gi ven by:
9942 . 0 ) )
100
20
( exp( ) ) ( exp( ) (
2 . 3
· − · − ·
β
η
A
A A
t
t R
1 )
32
240 400
( ) ( ) ( ·
−
Φ ·
−
Φ ·
σ
µ
B
B B
t
t R
8174 . 0 ) 1440 00015 . 0 exp( ) exp( ) ( · × − · × − ·
C C C
t t R λ
The syst em r el i abi l i t y f or 240 cycl es i s gi ven by:
8126 . 0 8174 . 0 1 9942 . 0 ) 1440 ( ) 240 ( ) 20 ( ) 240 ( · × × · × × ·
C B A S
R R R R
4.22. PARALLEL CONFIGURATION
In a par al l el conf i gur at i on t he syst em f ai l s onl y when al l t he i t ems of t he
syst em f ai l . In ot her w or ds, t o mai nt ai n t he r equi r ed f unct i on onl y on e i t em
4. Syst ems Rel i abi l i t y 106
of t he syst em i s r equi r ed t o f u nct i on. The r el i abi l i t y bl ock di agr am f or a
syst em consi st i ng of i t ems connect ed i n par al l el i s shown i n Fi gur e 4.7
Fi gur e 4.7 Rel i abi l i t y bl ock di agr am f or a par al l el conf i gur at i on
Par al l el component s ar e i nt r oduced when t he r el i abi l i t y r equi r em ent s f or
t he syst em ar e ver y hi gh. The use of mor e t han one engi ne i n ai r cr af t i s one
of t he obvi ous exampl es of par al l el conf i gur at i on (In pr act i ce an ai r cr af t
w oul d not be al l ow ed t o f l y i f any of t he engi ne f ai l s. I f an engi ne f ai l s
dur i ng a f l i ght , t he pi l ot woul d nor mal l y be expect ed t o di ver t t o t he
near est ai r por t ). How ever , par al l el i t ems w i l l i ncr ease cost , compl exi t y and
wei ght of t he syst em. Hence, t he number of par al l el it ems r equi r ed shoul d
be car ef ul l y det er mi n ed and i f possi bl e opt i mi sed.
Rel i abi l i t y f unct i on of par al l el conf i gurat i on
Rel i abi l i t y f unct i on of a par al l el conf i gur at i on can be ob t ai ned usi ng t he
f ol l owi ng ar gument s. As t he syst em f ai l s onl y wh en al l t he i t ems f ai l , t he
f ai l u r e f unct i on, F
S
(t ), of t he syst em i s gi ven by:
] ,... , [ ) (
2 1
t TTF t TTF t TTF P t F
n S
≤ ≤ ≤ · (4.11)
wher e TTF
i
r epr esent s t he t i met of ai l ur e r andom var i abl e of i t em i .
Assumi ng i ndependen ce among di f f er ent i t ems, t he above expr essi on can
be wr i t t en as:
F
S
(t ) = F
1
(t ) × F
2
(t ) × …× F
n
(t )
(4.12)
1
2
n
4. Syst ems Rel i abi l i t y 107
wher e F
i
(t ) i s t he t i me t o f ai l ur e di st r i but i on of i t em i . Subst i t ut i ng F
i
(t )=1 
R
i
(t ) i n equat i on (4. 12), t h e expr essi on f or f ai l ur e f unct i on of a par al l el
conf i gur at i on can be wr i t t en as:
F
S
(t ) = [ 1  R
1
(t ) ] × [ 1  R
2
(t ) ] × …× [ 1  R
n
(t ) ]
(4.13)
Now, t he r el i abi l i t y f unct i on, R
S
(t ), of a par al l el conf i gur at i on can be wr i t t en
as:
R
S
(t ) = 1  F
S
(t ) = 1  [ 1  R
1
(t )] × [ 1  R
2
(t )] × …× [ 1 R
n
(t )]
or
∏
·
− − ·
n
i
i S
t R t R
1
)] ( 1 [ 1 ) ( (4.14)
Charact eri st i cs of a paral l el conf i gurat i on
1. The syst em r el i abi l i t y, R
S
(t ), i s mor e t han r el i abi l i t y of t he any of t he
consi st i ng i t ems. That i s,
R
S
(t) ≥ )} ( {
,..., 1
t R Max
i
n i·
2. If h
i
(t ) r epr esent t h e hazar d r at e of i t em i , t hen t he r el i abi l i t y f unct i on of
a par al l el conf i gur at i on can be wr i t t en as:
∏ ∫
·
− − − ·
n
i
t
i S
dt t h t R
1 0
] ) ( exp( 1 [ 1 ) (
Exampl e 4.6
A f l ybywi r e ai r cr af t has f our f l i ght cont r ol syst em el ect r o ni cs (FCSE)
connect ed i n par al l el . The t i met of ai l ur e of FCSE can be r epr esent ed by
Wei bul l dist r i bu t i on wi t h scal e par amet er η=2800 and β = 2. 8. Fi nd t he
r el i abi l i t y of f l i ght cont r ol syst em f or 1000 hour s of oper at i on.
SOLUTION:
Rel i abi l i t y f unct i on f or a paral l el syst em wi t h f our i dent i cal i t ems i s gi ven by:
4. Syst ems Rel i abi l i t y 108
R t R t
R t
S i
i
( ) [ ( )]
[ ( )]
· − − ∏
· − −
·
1 1
1 1
1
4
4
wher e R(t ) i s t he r el i abi l i t y f unct i on of each i t em. For t = 1000, R(t ) i s gi ven
by:
9455 . 0 ) ) 2800 / 1000 ( exp( ) ) / ( exp( ) (
8 . 2
· − · − ·
β
η t t R
Thus t he r el i abi l i t y of f l i ght cont r ol syst em f or 1000 hour s of oper at i on i s
gi ven by:
999991 . 0 ] 9455 . 0 1 [ 1 ) 1000 (
4
· − − ·
S
R
Hazard f unct i on of a par al l el conf i gurat i on
Hazar d f unct i on, h
S
(t ), of t he par al l el conf i gur at i on can be wr i t t en as:
) (
1 ) (
) (
t R dt
t dR
t h
S
S
S
×
−
· (4.15)
Subst i t ut i ng t he expr essi on f or R
S
(t ) f r om equat i on (4.14) i n t he above
equat i on, we get
∏
∏
·
·
− −
× − − − ·
n
i
i
n
i
i S
t R
t R
dt
d
t h
1
1
))] ( 1 ( 1 [
1
} ))] ( 1 ( 1 [ { ) ( (4.16)
It i s easy t o ver i f y t hat t he above equat i on can be wr i t t en as:
∏
∑ ∏
·
· ≠ ·
− −
×
·
n
i
i
n
j
n
j i i
i j
S
t R
t F t f
t h
1
1 , 1
)] ( 1 [ 1
} ) ( ) ( {
) ( (4.17)
Wher e, f
i
(t ) i s t he pr obabi l i t y densi t y f unct i on of i t em i .
Exampl e 4.7
4. Syst ems Rel i abi l i t y 109
For t he f l i ght cont r ol syst em el ect r oni cs di scussed i n t he exampl e 3.5, f i nd
t he hazar d f unct i on of t he syst em at t i me t = 100.
SOLUTION:
Si nce al l t he f our i t ems ar e i dent i cal , t he hazar d r at e of t he syst em can be
w r i t t en as (usi n g equat i on (3.15)):
4
3
)] ( [ 1
)] ( [ ) ( 4
) (
t F
t F t f
t h
S
−
× ×
·
wher e,
) ) ( exp( ) (
) ) ( exp( ) ( ) (
1
β
β β
η
η η η
β
t
t F
t t
t f
− ·
− ·
−
Subst i t ut i ng t = 100, we get
h
S
(t ) = 8.0 × 10
 8
Mean t i me t o f ai l ure of paral l el conf i gurat i on
The mean t i me t o f ai l ur e of a par al l el conf i gur at i on, denot ed by M TTF
S
, can
be wr i t t en as:
∫ ∏ ∫
∞
·
∞
− − · ·
0 1 0
)]} ( 1 [ 1 { dt t R dt R MTTF
n
i
i S S
(4.18)
For most of t he f ai l ur e di st r i but i ons one may have t o use n umer i cal
i nt egr at i on t o eval uat e t h e abo ve i nt egr al . How ever , i n case of exponent i al
di st r i but i on we can get si mpl e expr essi on f or syst em’ s M TTF.
Assume t hat t he t i met of ai l ur e di st r i but i on of component i i s
expon ent i al wi t h mean ) / 1 (
i
λ . Then t he mean t i me t o f ai l ur e of t he
syst em, M TTF
S
, i s gi ven by:
4. Syst ems Rel i abi l i t y 110
∫ ∏ ∫ ∏
∞
·
∞
·
− − − · ·
0 1 0 1
)]} exp( 1 [ 1 { ) (
n
i
i
n
i
i s
dt t dt t R MTTF λ (4.19)
For par t i cul ar val ues of n, we can si mpl i f y t h e above i nt egr al t o d er i ve t he
expr essi on f or t he M TTF
S
.
Case 1: Assume n = 2. Equat i on (4.19) can be wr i t t en as:
2 1 2 1
0
2 1 2 1
2
0
1
1 1 1
)] ) ( exp( ) exp( ) [exp(
))]} exp( 1 ( )) exp( 1 [( 1 {
λ λ λ λ
λ λ λ λ
λ λ
+
− + ·
+ − − − + − ·
− − ⋅ − − − ·
∫
∫
∞
∞
dt t t t
dt t t MTTF
S
Case 2: Assume n = 3, t he expr essi on f or M TTF
S
can be wr i t t en as:
3 2 1 3 2 3 1 2 1 3 2 1
0
3
1
1 1 1 1 1 1 1
))]} exp( 1 ( [ [ 1 {
λ λ λ λ λ λ λ λ λ λ λ λ
λ
+ +
+
+
−
+
−
+
− + + ·
− − − ·
∫ ∏
∞
·
dt t MTTF
i
i
S
(4.20)
4.23. ROUTOFN SYSTEMS
In an r out of n (or r out of n: G) syst em, at l east r i t ems out of t he t ot al n
i t ems shoul d mai nt ai n t hei r r equi r ed f unct i on f or t he syst em t o be
oper at i onal . Fol l owi ng ar e f ew exampl es of r out of n syst ems:
1. Control software in a space shuttle has four programs. For the successful
completion of the mission, at least three of them should maintain the
required function and also the output from at least three programs should
agree with each other. This is an example of a 3outof4 system.
2. Most of the telecommunication system can be represented as a routofn
systems.
The r el i abi l i t y f unct i on of r out of n syst em can be der i ved as st at ed bel ow .
4. Syst ems Rel i abi l i t y 111
Rel i abi l i t y f unct i on of an rout of n syst em
Consi der an r out of n syst em wi t h i dent i cal i t ems. That i s, R
1
(t )=R
2
(t ) =. .. =
R
n
(t ). Then t he syst em r el i abi l i t y, R
S
(t ,r ,n), i s gi ven by:
i n i
n
r i
S
t R t R
i
n
n r t R
−
·
−
,
_
¸
¸
·
∑
)] ( 1 [ )] ( [ ) , , ( (4.21)
For t he cases when t he t i met of ai l ur e di st r i but i on i s exponent i al or Wei bul l
w e have t he f ol l ow i ng expr essi ons f or r el i abi l i t y f unct i on.
1. Exponent i al t i met of ai l ur e di st r i but i on
∑
·
−
− − −
,
_
¸
¸
·
n
r i
i n i
S
t t
i
n
n r t R ] exp( 1 [ )] [exp( ) , , ( λ λ
2. Wei bul l t i met of ai l ur e di st r i but i on
∑
·
−
− − −
,
_
¸
¸
·
n
r i
i n i
S
t t
i
n
n r t R )] ) ( exp( 1 [ )] ) ( [exp( ) , , (
β β
η η
However , i f t he i t ems ar e not i dent i cal t hen one may have t o use ot her
mat h emat i cal model s such as enumer at i on t o eval uat e t h e r el i abi l i t y. For
exampl e consi der a 2ou t of 3 syst em wi t h noni dent i cal i t ems. The
r el i abi l i t y f unct i on of t he syst em can be d er i ved as f ol lows.
Let E
i
denot e t he event t hat t he i t em i successf u l l y compl et es t he mi ssi on
(or sur vi ves t hour s of oper at i on). Th en t he r el i abi l i t y f unct i on f or t he
syst em can be wr i t t en as:
R
S
(t ) = P [ {E
1
∩ E
2
} ∪ {E
1
∩ E
3
} ∪ {E
2
∩ E
3
} ]
By put t i ng, A = E
1
∩ E
2
, B = E
1
∩ E
3
and C = E
2
∩ E
3
, t he above expr essi on
can be w r i t t en as:
R
S
(t ,2,3) = P [ {A ∪ B ∪ C } ]
= P(A) + P(B) + P(C)  P(A ∩ B)  P(A ∩ C)  P(B ∩ C)
+ P(A ∩ B ∩ C)
4. Syst ems Rel i abi l i t y 112
= P(E
1
∩E
2
) + P(E
1
∩ E
3
) + P(E
2
∩ E
3
)  2 P( E
1
∩ E
2
∩E
3
)
Let R
i
( t ) r epr esent t he r el i abi l i t y f unct i on f or t he i t em i . Now t he abo ve
expr essi on can be wr i t t en as:
R
S
(t , 2, 3) = R
1
(t ) R
2
(t ) + R
1
(t ) R
3
(t ) + R
2
(t ) R
3
(t )  2 × R
1
(t ) R
2
(t ) R
3
(t )
The abo ve appr oach becomes compl ex wh en t he number of i t ems n
i ncr eases. How ever , t her e ar e sever al appr oaches avai l abl e t o t ackl e
compl ex r out of n syst ems wi t h noni dent i cal i t ems. The r el i abi l i t y f unct i on
of r 1out of n and r out of n syst em wi t h i dent i cal i t ems sat i sf i es t he
f ol l ow i ng r el at i on:
) , , ( )] ( 1 [ )] ( [
1
) , 1 , (
1 1
n r t R t R t R
r
n
n r t R
S
r n r
s
+ −
,
_
¸
¸
−
· −
+ − −
(4.22)
Mean Ti me t o Fai l ure of rout of n Syst ems
The mean t i me t o f ai l ur e, M TTF, of an r out of n syst em, M TTF
S
(r ,n), can be
obt ai ned usi ng t he f ol l owi ng expr essi on:
∫
∞
·
0
) , , ( ) , ( dt n r t R n r MTTF
S S
One may have t o use numer i cal i nt egr at i on i n most of t he cases t o eval uat e
t he abo ve i nt egr al . How ever , i f t he t i met of ai l ur e di st r i but i on i s
expon ent i al , t hen t he above i nt egr al r educes t o a si mpl e expr essi on. For
exampl e, consi der a 2ou t of 3 syst em wi t h i dent i cal i t ems wher e t h e t i me
t of ai l ur e di st r i but i on of t h e i t em i s r epr esent ed b y exponent i al di st r i but i on
wi t h par amet er λ. The r el i abi l i t y f unct i on of 2o ut of 3 syst em wi t h
expon ent i al i t ems ar e gi ven by:
) 3 exp( )) exp( 1 )( 2 exp( 3
)] exp( 1 [ )] [exp(
3
) (
3
2
t t t
t t
i
t R
i
i n i
S
λ λ λ
λ λ
− + − − − ·
− − −
,
_
¸
¸
·
∑
·
−
Now t he M TTF
S
i s gi ven by,
4. Syst ems Rel i abi l i t y 113
MTTF t t t dt
S
· − − − + −
∫
·
∞
[ exp( )( exp( )) exp( ] 3 2 1 3
5
6
0
λ λ λ
λ
Usi ng equat i on (4.22), we get t he f ol l owi ng r el at i on bet ween M TTF
S
(r 1,n)
and M TTF
S
(r ,n) (M i sr a, 1992):
) , ( )] ( 1 [ )] ( [
1
) , 1 (
0
1 1
n r MTTF dt t R t R
r
n
n r MTTF
s
r n r
S
+ −
,
_
¸
¸
−
· −
∫
∞
+ − −
(4.23)
4.24. SERIES AND PARALLEL CONFIGURATION
In t hi s Sect i on w e di scuss t w o t ypes of ser i es and par al l el st r uct ur es, w hi ch
have w i de appl i cat i on i n r el i abi l i t y t heor y.
Model 1. Ser i esPar al l el Conf i gur at i on
Her e t he syst em has a ser i es st r uct ur e wi t h n i t ems wher e each i t em has
par al l el r edundant i t ems. Assume t hat i t em i has m
i
component s i n par al l el .
Fi gur e 4.8 shows a ser i espar al l el conf i gur at i on.
Fi gur e 4.8 Ser i espar al l el st r uct ur e w i t h n i t ems subsyst em wher e
subsyst em i has m
i
par al l el component s
In Fi gur e 4.8, (i ,j ) r epr esent j t h par al l el component of t he i t em i . If R
i, j
(t )
denot e t h e cor r espondi ng r el i abi l i t y of t he co mponent , t hen t he r el i abi l i t y
of i t em i of t he syst em i s gi ven by:
∏
·
− − ·
i
m
j
j i i
t R t R
1
,
)] ( 1 [ 1 ) ( (4.24)
1,1
1,2
1,m1
2,1
2,2
2,m2
n,1
n,2
n,mn
4. Syst ems Rel i abi l i t y 114
Now t he syst em r el i abi l i t y can be w ri t t en as:
∏ ∏ ∏
· · ·
− − · ·
n
i
m
j
j i
n
i
i S
i
t R t R t R
1 1
,
1
))] ( 1 ( 1 [ ) ( ) ( (4.25)
Model 2. Par al l el Seri es Syst em
Fi gur e 4. 9. Par al l el ser i es st r uct ur e wi t h n subsyst ems wher e sub syst em i
has m
i
componen t s
Assume t hat t h e syst em has n i t ems connect ed i n par al l el w her e each i t em
has component s conn ect ed i n ser i es. An ai r cr af t wi t h mor e t han one
engi n e, i s a t ypi cal exampl e f or t hese t ype of conf i gur at i on. Fi gur e 4.9
shows par al l el ser i es st r uct ur e.
Si nce i t em i has m
i
component s i n ser i es, t he r el i abi l i t y of i t em i i s gi ven by:
∏
·
·
i
m
j
j i i
t R t R
1
,
) ( ) ( (4.26)
wher e R
i, j
(t ) i s t he r el i abil i t y f unct i on of t he compon en t j i n i t em i . Now t he
r el i abi li t y of t he par al l el ser i es syst em i s gi ven by:
∏ ∏ ∏
· · ·
− − · − − ·
n
i
n
j
j i
n
i
i S
t R t R t R
1 1
,
1
] ) ( 1 [ 1 )] ( 1 [ 1 ) ( (4.27)
1,1
2,1
n,1
1,2
2,2
n,2
1,m1
2,m2
n,mn
.
.
.
.
.
.
.
.
.
.
.
.
4. Syst ems Rel i abi l i t y 115
4.25. REDUNDANT SYSTEMS
In syst ems, r edundancy i s a means of mai nt ai ni ng syst em i nt egr i t y i f
cr i t i cal par t s of i t f ai l . I n some cases t hi s means r epl i cat i ng par t s of t he
syst em, i n ot her s, al t er nat i ves ar e used. A commer ci al ai r cr af t has t o be
abl e t o compl et e a t akeof f and l andi ng wi t h on e of i t s engi nes shut down
but , except under ver y speci al ci r cumst ances, no such ai r cr af t w oul d be
al l ow ed t o l eave t he depar t ur e gat e i f any of i t s engi nes ar e not f unct i oni ng.
And yet , ETOPS, ext end ed t wi n engi ne oper at i ons al l ows cer t i f i ed t wi n
engi n e ai r cr af t (e.g. Boei ng 777 and Ai r bus 330) t o f l y up t o 180 mi nut es
f r om a sui t abl e l andi ng si t e. Thi s i s based on t h e pr obabi l i t y t hat even i f
one of t he engi nes f ai l s t hat f ar f r om l and, t he o t her i s suf f i ci ent l y, r el i abl e
t o make t he pr obabi l i t y of not r eachi ng a l andi ng si t e an accept abl e r i sk. I t
shoul d be not ed t hat i n nor mal f l i ght , i .e. at cr ui si ng speed and al t i t ude, t he
engi n es ar e gener al l y doi ng ver y l i t t l e wor k and usual l y ar e t hr o t t l ed back.
If an engi ne f ai l s, i t woul d nor mal l y be wi ndmi l l ed t o mi ni mi se ' par asi t i c'
dr ag but , even t hen, i t st i l l of f er s a consi der abl e r esi st ance and, of cour se,
pr oduces an i nbal ance whi ch has t o b e of f set by t he r udd er and ot her
cont r ol l abl e sur f aces al l of whi ch means t he f unct i onal engi ne has t o w or k
consi der abl y har der t hus i ncr easi ng i t s pr obabi l i t y of f ai l ur e.
If t he ai r cr af t onl y had one engi n e and i t f ai l ed, t h e pr obabi l i t y of l andi ng
saf el y wi t h no en gi nes i s not ver y hi gh, at l east , f or f ast mi l i t ar y j et s. In
4. Syst ems Rel i abi l i t y 116
most cases ul t i mat el y, i f t he engi ne cannot be r el i t , t he onl y opt i on i s t o
ej ect af t er di r ect i ng t he ai r cr af t away f r o m i nhabi t ed ar eas, i f t her e i s t i me.
Wi t h commer ci al ai r l i nes, nei t her t h e pi l ot , t he cr ew nor t h e passenger s
have t he op t i on of ej ect i ng or bal i ng o ut i f t he ai r cr af t suf f er s a t ot al engi ne
f ai l ur e (i .e. al l engines f ai l ). These ai r cr af t w i l l gli de, t o a cer t ai n ext end b ut ,
wi t h no power , none of t h e i nst r ument s wi l l f unct i on and, t her e wi l l be no
pow er assi st ance f or t he cont r ol sur f aces or t o depl oy t he l andi ng gear . For
t hi s r eason, t hey ar e f i t t ed wi t h wi nd t ur bi nes t hat shoul d dr op down and
st ar t f unct i oni ng i f t her e i s pr ol onged l oss of power . Thi s gi ves t he pi l ot s
some cont r ol , but even t hen, l ar ge ai r l i ner s ar e not goi ng t o r i se on a
t her mal , how ever go od t he pi l ot may be.
A Boei ng 767, on one of i t s f i r st f l i ght s, had a t ot al engi ne f ai l ur e some
1500 mi l es f r om i t s i nt end ed dest i nat i on, Ont ar i o. Al l at t empt s t o r el i ght
t he engi nes f ai l ed si mpl y because i t had r un out of f u el . Th er e was a t o t al
bl ackout i n t he cockpi t and, even when t h e copi l ot managed t o f i nd a t or ch
(f l ashl i ght s) al l t hi s showed was t hat none of t he i nst r ument s wer e wor ki ng
(bei ng al l di gi t al and comput er cont r ol l ed). The pi l ot , by pur e chance,
happened t o be an ext r emel y accompl i shed gl i der pi l ot and, agai n by pur e
chance, t he copi l ot happened t o be par t i cul ar l y f ami l i ar wi t h t hi s par t of
Canada, some 200 mi l es out si d e Wi nni peg. For sever al mi nut es t he pi l ot
manhandl ed t he cont r ol s and managed t o st op t he ai r cr af t f r om l oosi ng
hei ght t oo qui ckl y. Event ual l y t he wi nd t ur bi ne depl o yed whi ch gave t hem
enough pow er f or t he i nst r ument s, r adi o and pow er assi st ed cont r ol s t o
wor k agai n. Unf or t unat el y t he ai r cr af t had l ost t o o much hei ght t o r each
Wi nni peg but , i t had j ust enough t o get t o an ex mi l i t ar y r unw ay (used as a
st r i p f or dr ag r aci ng). Ther e was j ust enough power t o l ock t he mai n
under car r i age dow n, but not t h e nose w heel . Th e Gi ml i Gl i der as i t became
know n, l anded saf el y w i t h no ser i ous casual t i es. But , out of el even ot her
pi l ot s, w ho l at er t r i ed t o l and t h e ai r cr af t i n t he same ci r cumst ances on a
f l i ght si mul at or al l cr ashed. Had i t not been f or t he ' r edundant ' wi nd
t ur bi ne, i t i s al most cer t ai n even t hi s exper i enced gl i der pi l ot woul d have
cr ashed ki l l i ng al l on boar d.
If t he Bo ei ng 777, say, was f i t t ed wi t h t hr ee or f our Rol l s Ro yce Tr ent
800s, Pr at t & Whi t ney 4084s or Gener al El ect r i c GE 90' s (i nst ead of t he t wo
i t cur r ent l y has) t hen t her e woul d be t r u e r edundan cy si nce i t needs onl y
t w o t o achi eve ETOPS (Ext ended Twi nengi ne Oper at i ons). Ther e ar e,
however , a nu mber of pr obl ems wi t h t hi s desi gn. Fi r st l y, i t woul d add ver y
si gni f i cant l y t o bot h wei ght and dr ag, t o t he poi nt wher e i t woul d ser i ousl y
r educe t h e payl oad and r ange, pr obabl y maki ng t h e ai r cr af t unecono mi cal
t o oper at e and hence undesi r abl e t o t he ai r l i nes. Secondl y such an i ncr ease
i n wei ght and dr ag woul d pr obabl y mean t he nor mal t wo engi nes woul d
4. Syst ems Rel i abi l i t y 117
pr ovi de i nsuf f i ci ent t hr ust t her ef or e ei t h er mor e power f ul en gi nes woul d
be n eeded or , t h e ext r a engi nes w oul d h ave t o be used r ender i ng t hem no
l onger t r ul y r edun dant .
On t he Bo ei ng 767, f or exampl e, t he IFSD (In Fl i ght Shut Down) r at e
af t er 10 mi l l i on hour s w as l ess t han 0.02 per t housand f l yi ng hour s (t he
st andar d measur e i n t he aer ospace i ndust r y). And, none of t hese had l ed
t o t he l oss of a si ngl e l i f e, l et al on e an ai r cr af t w i t h i t s f ul l compl ement of
passenger s and cr ew. It i s qui t e l i kel y t hat , i n some of t he i nst ances, f l i ght s
woul d have been di ver t ed f r om t h ei r schedul ed d est i nat i ons t o al t er nat i ves,
f or saf et y r easons. Th e i nconveni en ce t o passenger s (and ai r l i nes) woul d
have cost t he ai r l i ne but , t he amount woul d, al most cer t ai nl y, have been
si gni f i cant l y l ess t han t he l oss of r evenue r esul t i n g f r om t h e r educed
payl oad had t r ul y r edundant engi nes been f i t t ed.
In many cases, t he r edu ndant i t ems may not be f unct i oni ng si mul t aneousl y
as i n t he case of par al l el or r out of n conf i gur at i ons. The r edundant i t ems
wi l l be t ur ned on onl y when t he mai n i t em f ai l s. In some cases, t he i t ems
may b e f unct i oni ng si mul t aneousl y but one of t hem may be shar i ng much
hi gher l oad compar ed t o t h e ot her . Such t yp es of syst ems ar e cal l ed
st andby r edundant syst ems. Whenever t he mai n i t em f ai l s, a bui l t i n swi t ch
senses t h e f ai l ur e and swi t ches on t he f i r st st andb y i t em. It i s i mpor t an t
t hat t he sw i t ch has t o mai nt ai n i t s f unct ion. Fai l ur e of t he sw i t ch can cause
t he syst em f ai l ur e. The st andb y r edundant syst ems ar e nor mal l y cl assi f i ed
as col d st andby, w ar m st andby and hot st andby.
Col d St andby Syst em
In a col d st andby, t he r edundant par t of t he syst em i s swi t ched on onl y
when t he mai n par t f ai l s. For exampl e, t o meet t h e const ant l y changi ng
demand f or el ect r i ci t y f r om t he ' Nat i onal Gr i d' i t i s necessar y t o keep a
number of st eam t ur bi nes r eady t o come on st r eam w h en ever t h er e i s a
sur ge i n demand. Th e f ai l ur e of a gener at or woul d r esul t i n i nst ant aneous
r educt i on i n capaci t y, whi ch woul d be r ect i f i ed by br i ngi ng one of t hese
' r edundant ' t ur bi nes up t o f ul l power . I n t he event of a power cut t o a
hospi t al , bat t er i es may swi t ch i n i nst ant l y t o pr ovi de emer gen cy l i ght i ng
and keep emer gency equi pment , e. g. r espi r at or s and moni t or s r unni n g.
Pet r ol and di esel gen er at or s woul d t hen be st ar t ed up t o r el i eve t he
bat t er i es and pr ovi de addi t i onal pow er .
In a col d st andby syst em, a r edundant i t em i s swi t ched on onl y when t he
oper at i ng i t em f ai l s. That i s, i ni t i al l y one i t em wi l l be oper at i ng and wh en
t hi s i t em f ai l s, one i t em f r om t he r edundan t i t ems w i l l be sw i t ched on t o
4. Syst ems Rel i abi l i t y 118
mai nt ai n t he f unct i on. In a col d st andby, t he hazar d f unct i on of t he i t em i n
st andby mode i s zer o.
Fi gur e 4.10 Col d st andby r edundan t syst em
Consi der a col d st andby syst em wi t h t wo i dent i cal i t ems (see Fi gur e 4.10).
The r el i abi l i t y f unct i on of t hi s syst em can b e der i ved as f ol l ow s (assumi ng
t hat t he swi t ch i s per f ect ):
R
S
(t ) = P{The mai n i t em sur vi ves up t o t i me t }
+ P{ Th e mai n i t em f ai l s at t i me u ( u < t ) and t he st andby
i t ems sur vi ves t he r emai ni ng i nt er val ( t  u ) }
Thus,
du u t R u f t R t R
t
S ∫
− + ·
0
) ( ) ( ) ( ) ( (4.28)
wher e f (t ) i s t he pr obabi l i t y densi t y f un ct i on of t i met of ai l ur e r andom
var i abl e.
As an exampl e consi d er a col d st andb y syst em wi t h t wo i t ems wher e t he
t i met of ai l ur e di st r i but i on i s exponent i al w i t h par amet er λ. Usi ng t he
equat i on (4.29) t he expr essi on f or r el i abil i t y f unct i on i s gi ven by:
] 1 )[ exp( ) exp( ) exp(
)) ( exp( ) exp( ) exp( ) (
0
t t t t t
du u t u t t R
t
S
λ λ λ λ λ
λ λ λ λ
+ − · − + − ·
− − × − + − ·
∫
For a col d st andby syst em wi t h n i dent i cal i t ems wi t h exponent i al t i met o
f ai l ur e di st r i but i on, t he expr essi on f or r el i abi l i t y f unct ion i s gi ven by:
1
2
4. Syst ems Rel i abi l i t y 119
∑
−
·
− ·
1
0
!
) (
) exp( ) (
n
i
i
S
i
t
t t R
λ
λ (4.29)
The equat i on (4.30) i s t h e cu mul at i ve di st r i but i on of Poi sson di st r i but i on
wi t h mean λt . One can al so der i ve t h e expr essi on f or noni dent i cal st andb y
uni t s usi ng t he ar gu ment s pr esent ed i n equat i on (4.29). For a col dst andby
syst em wi t h noni dent i cal i t ems, t he syst em r el i abi l i t y f unct i on i s gi ven by:
dx x t R x f t R R
t
s ∫
− + ·
0
2 1 1
) ( ) ( ) ( (4.30)
Wh er e R
1
(t ) and f
1
(t ) ar e t he r el i abi li t y f unct i on and f ai l ur e densi t y f unct i on
of it em 1 and R
2
(t ) i s t he r el i abi li t y f unct i on of i t em 2. Assume t hat t he
t i met of ai l ur e i t ems 1 and 2 can be model l ed usi ng exponent i al
di st r i but i on wi t h mean (1/ λ
1
) and (1/ λ
2
) r espect i vel y. Usi ng equat i on
(4.31), t he r el i abi l i t y f unct i on of col dst andby syst em wi t h noni dent i cal
i t ems i s wi t h exponent i al f ai l ur e t i me i s gi ven by:
∫
− − × − + − ·
t
s
dx x t x t t R
0
2 1 1 1
)) ( exp( ) exp( ) exp( ) ( λ λ λ λ
)] exp( ) [exp( ) ( exp( ) (
1 2
2 1
1
1
t t t t R
s
λ λ
λ λ
λ
λ − − −
−
+ − ·
The M TTF of a col dst andby syst em can be eval uat ed by i nt egr at i n g t he
r el i abi l i t y f unct i on b et w een 0 and ∞. Th e M TTF of a col dst andby syst em
wi t h n i dent i cal uni t s wi t h exponen t i al f ai l ur e t i me i s gi ven by:
λ
n
MTTF · (4.31)
Equat i on (4.31) can be easi l y der i ved f r om equat i on (4.30). For t he non
i dent i cal M TTF i s gi ven by:
∑ ·
·
n
i
i
s
MTTF
1
1
λ
(4.31a)
Warm St andby Syst em
4. Syst ems Rel i abi l i t y 120
In a w ar m st andby syst em, t h e r ed undant i t em w i l l be shar i ng par t i al l oad
al ong w i t h t h e mai n i t em. Thus, i n a w ar m st andb y, t he hazar d f unct i on of
t he st andb y i t em wi l l be l ess t han t hat of t he mai n i t em.
That i s, a st andby syst em can det er i or at e even when i t i s not i n use.
Consi der a syst em wi t h t wo war m st andby i t ems. Assume t hat R(t ) and
R t
s
( ) r epr esent t he r el i abi l i t y of t he i t em i n oper at i ng mode and st and by
mod e r espect i vel y. Now t he r el i abi l i t y f unct i on of t he syst em can be w r i t t en
as:
du x t R x R x f t R t R
t
s
S ∫
− × × + ·
0
) ( ) ( ) ( ) ( ) ( (4.32)
For a par t i cul ar case wher e ) exp( ) ( t t R λ − · and ) exp( ) ( t t R
s
s
λ − · t he
r el i abi l i t y f unct i on of a war m st andby syst em i s gi ven by:
)) exp( 1 (
) exp(
) exp(
)) ( exp( ) exp( ) exp( ) exp( ) (
0
t
t
t
du u t u u t t R
s
s
t
s S
λ
λ
λ λ
λ
λ λ λ λ λ
− −
−
+ − ·
− − × − × − + − ·
∫
Hot St andby Syst em
In a hot st andby, t he mai n i t em and t he st and by i t em wi ll be shar i n g equal
l oad, and hence wi l l have t he same hazar d r at e. Thus, a hot st andby can be
t r eat ed as a par all el syst em t o d er i ve r el i abi l i t y expr essi ons. If h
o
(t ) and h
s
(t )
r epr esen t t he hazar d r at e of a oper at i ng and st andby i t em r espect i vel y.
The Tabl e 4.6 gi ves t h e var i ous r edundanci es and t he pr oper t i es of hazar d
r at e.
Tabl e 4.6 Types of st andby r edundancy and t he cor r espondi ng pr oper t i es of
hazar d r at e
Type of Redundancy Pr op er t i es of hazar d r at e
4. Syst ems Rel i abi l i t y 121
Col d St andby h
S
(t ) = 0
War m St andb y h
O
(t ) > h
S
(t )
Hot St andby h
O
(t ) = h
S
(t )
4.26. COMPLEX RELIABILITYBLOCK DIAGRAMS
In many cases, t he r el i abi l i t y bl ock di agr am w i l l have co mpl ex
combi nat i ons of ser i es and par al l el b l ocks. In such cases, one has t o r educe
t he bl ock t o ei t her a ser i es st r uct ur e or a par al l el st r uct ur e b ef or e one can
pr edi ct t he r el i abi l i t y char act er i st i cs of t he syst em. Red uci ng a co mpl ex
r el i abi l i t y st r uct ur e wi l l i nvol ve t he f ol l owi ng st eps:
1. Replace all purely series (parallel) with an equivalent (reliability wise)
single block.
2. Repeat step 1 up till the RBD reduces to either a series or parallel
structure.
3. Compute the reliability of resulting RBD.
For exampl e, consi der t he RBD shown i n Fi gur e 4. 11.
Figure 4.11 Reliability block diagram with combination of seriesparallel
structures
The t i met of ai l ur e of t he si x i t ems wi t hi n t he syst em shown i n Fi gur e 4.11
ar e show n i n Tabl e 4.7.
Tabl e 4.7. Ti met o f ai l u re o f i t ems show n in Fi gur e 4.12
It em Di st r i but i on wi t h par amet er val ues
2 3
4
5
6
1
4. Syst ems Rel i abi l i t y 122
1 Wei bul l , η = 450 hour s, β = 2.4
2 Lognor mal µ
l
= 4.5, σ
l
= 0.75
3 Wei bul l , η = 890 hour s, β = 1. 75
4 Exponent i al , λ = 0.001
5 Nor mal µ = 800, σ = 120
6 Exponen t ial , λ = 0. 00125
The r el i abi l i t y bl o ck di agr am shown i n Fi gur e 4.11 can be eval uat ed usi ng
t he t hr ee st eps expl ai ned above. The RBD i n Fi gur e 4. 11 can be r epl aced by
a ser i es st r uct ur e w i t h t hr ee bl ocks as show n i n Fi gur e 4. 11a.
Fi gur e 4.11a Rel i abi l i t y bl ock di agr am equi val ent t o Fi gur e 4.11
Fi gur e 4.11b RBD equi val ent t o bl ock B i n Fi gur e 4. 11
The bl ock A i s same as i t em 1, wher e bl ock B i s equi val ent t o t h e RBD
shown i n Fi gur e 4.11b.
The bl ock B i s equi val ent t o RBD shown i n Fi gur e 4.12.c.
A
B C
2 3
4
4
5
4. Syst ems Rel i abi l i t y 123
Fi gur e 4.11c. RBD equi val ent t o bl ock C i n Fi gur e 4.11
The expr essi on f or r el i abi l i t y f unct i on of t he syst em i n Fi gur e 4.11 i s gi ven
by:
) ( ) ( ) ( ) ( t R t R t R t R
C B A s
× × ·
wher e
) ( ) (
1
t R t R
A
·
))] ( 1 ( )) ( ) ( 1 ( 1 [ 1 ) (
4 3 2
t R t R t R t R
B
− × × − − − ·
))] ( 1 ( )) ( 1 ( 1 [ 1 ) (
6 5
t R t R t R
C
− × − − − ·
For some syst ems, t he r el i abi l i t y bl ock di agr am may have mor e compl ex
conf i gur at i on t han t he ser i es/ par al l el st r uct ur e as di scussed so f ar . The
w el l know n ‘ Wheat st one Br i dge’ (see Fi gur e 4.12) i s an exampl e of such
conf i gur at i on. To f i nd t he r el i abi l i t y of such syst ems one may have t o use
speci al t ool s such as cut set , pat hset , enumer at i on or t he condi t i onal
pr obabi l i t y appr oach. I n t hi s Sect i on w e i l l ust r at e t he cut set appr oach f or
eval uat i n g r el i abi l i t y of compl ex st r uct ur es.
4.27. CUT SET APPROACH FOR RELIABILITY
EVALUATION
Cut set appr oach i s one of t he most popul ar and wi del y used met hods f or
pr edi ct i ng r el i abi l i t y of compl ex st r uct ur e. The mai n advant age of cut set
appr oach i s t hat i t i s easy t o pr ogr am and most of t h e commer ci al sof t war e
f or r el i abi l i t y pr edi ct i on use cut set appr oach t o eval uat e t h e r el i abi l i t y of
compl ex st r uct ur es. A cut set i s def i ned as t he set of i t ems t hat , when
f ai l ed, wi l l cause t he syst em f ai l ure. A cut set wi t h mi ni mum number of
i t ems i s cal l ed mi ni mal cut set . That i s i f any i t em of t he mi ni mal cut set
has not f ai l ed, t hen t he syst em w i l l not f ai l . M at hemat i cal l y, i f t he set Ci s a
cut set of t he syst em. Then, t he set C w i ll be a mi ni mal cut set i f f or all c
i
∈
C, C  c
i
i s not a cut set . Her e C  c
i
r epr esent s t he set C w i t hout t he
el ement c
i
. The cut set appr oach t o r el i abi l i t y pr edi ct i on i nvol ves
i dent i f yi ng al l t he mi ni mal cut set s of t he syst em.
4. Syst ems Rel i abi l i t y 124
Fi gu r e 4.12 Br i d ge net w or k
In Fi gur e 4.12, t he set of i t ems C = { 1, 2, 3} f or ms a cut set , si nce t he f ail ur e
of t h e i t ems 1, 2 and 3 w i l l cause syst em f ai l ur e. How ever , t h e set C = {1, 2,
3} i s not a mi ni mal  cut set si nce C  3 = {1, 2} st i l l f orms a cut set . For t he
st r uct ur e show n i n Fi gur e 4. 12, t he mi ni mal cut set s ar e gi ven by:
C
1
= {1, 2}, C
2
= {1, 3, 5}, C
3
= {2, 3, 4} and C
4
= {4, 5}
Si nce al l t he el ement s of t he mi ni mal cut set shoul d f ai l t o cause t he syst em
f ai l ur e, each cut set can be consi der ed as a par al l el conf i gur at i on. Thus, t he
cut set s C
1
, C
2
, C
3
and C
4
r epr esent t h e f ol l owi ng st r uct ur es show n i n Fi gur e
4.13.
Fi gur e 4. 13. Equi val ent RBD f or mi ni mal cut set s of t he syst em show n i n
Fi gur e 4.12
Si nce t he syst em wi l l f ai l when at l east one mi ni mal cut set s f ai l , t he
r el i abi li t y f unct i on of t he syst em can be wr i t t en as:
5
1
2
3
4
1
2
1
5
3
2
3
4
4
5
C
1
=
C
C
2
= C
4
=
C
3
=
4. Syst ems Rel i abi l i t y 125
R
S
(t ) = RC
1
(t ) × RC
2
(t ) × RC
3
(t ) × RC
4
(t )
(4.33)
wher e RC
1
(t ), RC
2
(t ), RC
3
(t ) and RC
4
(t ) ar e t he r el iabi l i t y f unct i on of t he
st r uct ur es r epr esent ed b y t he cu t set s C
1
, C
2
, C
3
and C
4
r espect i vel y. If R
i
(t )
denot e t h e f ai l ur e f unct i on of t he i t ems 1, 2, 3, 4 and 5, t hen we have:
) ( ) ( 1 ) ( ), ( ) ( ) ( 1 ) (
) ( ) ( ) ( 1 ) ( ), ( ) ( 1 ) (
5 4 4 4 3 2 3
5 3 1 2 2 1 1
t F t F t RC t F t F t F t RC
t F t F t F t RC t F t F t RC
− · − ·
− · − ·
Subst i t ut i ng t he above expr essi ons i n equat i on (4.34), we get t he f ai l ur e
f unct i on f or t he compl ex st r uct ur e shown i n Fi gur e 4.12.
In gener al , cut set appr oach i nvol ves t he f ol l ow i ng st eps:
1. Identify all the minimal cut sets of the system.
2. Since all the elements of the minimal cut set should fail to cause the
system failure, each cut set can be treated as a parallel configuration.
3. Since failure of any one minimal cut set can cause system failure,
different minimal cut sets can be treated as a series configuration.
4. Syst ems Rel i abi l i t y 126
4.
4.28. CASE STUDY ON AIRCRAFT ENGINES
Ai r cr af t engi ne i s one of t he most cr i t i cal i t ems used i n t oday’ s avi at i on
i ndust r y. In t hi s sect i on, w e t r y t o addr ess sever al r el i abi l i t y measur es one
may l i ke t o know about an engi n e. Ther e ar e t o t al l y el even i t ems i ncl udi ng
t he ext er nal gear box, oi l t ank and f i l t er . Th e t i met of ai l ur e of t hese i t ems
ar e gi ven i n Tabl e 4.8.
Tabl e 4.8. Ti met of ai l ur e di st r i but i on of var i ous i t ems of t he engi ne
I t em
no.
It em Di st r i but i on Par amet er Val ues
01 LP compr essor Wei bul l η = 15 000, β = 3
02 LP st age 2 st at or Wei bul l η = 5 000, β = 2. 8
03 Int er medi at e
casi ng
Wei bul l η = 11 000, β = 3
04 HP compr essor Wei bul l η = 12 000, β = 3.5
05 HP NGV Wei bul l η = 8 000, β = 3
06 HP t ur bi ne Wei bul l η = 25 000, β = 4
07 LP NGV Wei bul l η = 7 000, β = 2. 2
08 LP t ur bi ne Wei bul l η = 20 000, β = 2.8
09 Exhaust mi xer Wei bul l η = 7 000, β = 3
10 Ext er nal gear box Wei bul l η = 6 500, β = 3
11 Oi l t ank and f i l t er Wei bul l η = 5 000, β = 3. 8
We ar e i nt er est ed i n car r yi ng out t he f ol l ow i ng t asks
1. Dr aw t he r el i abi l i t y bl ock di agr am of t he engi ne.
2. Fi nd r el i abi li t y of t he engi ne f or 3000 hour s of oper at i on.
4. Syst ems Rel i abi l i t y 127
3. Fi nd t he hazar d r at e of t he engi n e at t = 3000 and t = 7000 hour s.
4. Fi nd t he M TTF of di f f er ent i t ems of t he engi ne and est i mat e t he M TTF of
t he engi ne f r o m t he M TTF val u es of t he i t ems.
5. Fi nd t he M TTF of t h e engi ne i f al l t he i t ems ar e subj ect t o pr event i ve
mai nt enance af t er ever y 1000 hour s of op er at i on (assume t hat af t er
mai nt enance al l t he i t ems behave as good as new).
6. For an engi n e of age 5000 hour s, f i nd t he mi ssi on r el i abi l i t y f or 1000
hour s of oper at i on.
7. Fi nd t he M FOPS of t he engi ne f or 500 hour s of oper at i on f or di f f er en t
cycl es.
SOLUTION:
1. Si nce al l t h e i t em of t he engi ne must mai nt ai n t hei r f unct i on, t he syst em
w i l l have a ser i es conf i gur at i on as show n bel ow :
Fi gur e 4.14 Rel i abi l i t y bl ock di agr am of t he engi ne
2. Si nce al l t he i t ems of t h e syst em f ol l ow Wei bul l di st r i but i on, t he
r el i abi l i t y f unct i on f or each of t hese i t ems i s gi ven by:
) ) ( exp( ) (
β
η
t
t R − ·
Subst i t ut i ng t he val ues of η and β f or var i ous i t ems i n t he above equat i on,
t he r el i abi l i t y of var i ous i t ems f or 3000 hour s of oper at i on i s gi ven by:
1. Rel i abi l i t y of LP compr essor f or 3000 hour s of oper at i on i s gi ven by:
LP compressor
LP stage 2
stator
Intermediate
casing
HP compressor
HP NGV HP turbine LP NGV
LP Turbine
Exhaust mixer
External
gearbox
Oil tank and
filter
4. Syst ems Rel i abi l i t y 128
9920 . 0 ) )
15000
3000
( exp( ) 3000 (
3
1
· − · R
2. Rel i abi l i t y of LP st age 2 st at or f or 3000 hour s of oper at i on i s gi ven by:
7872 . 0 ) )
5000
3000
( exp( ) 3000 (
8 . 2
2
· − · R
3. Rel i abi l i t y of i nt er medi at e casi ng f or 3000 hour s of oper at i on i s gi ven by:
9799 . 0 ) )
11000
3000
( exp( ) 3000 (
3
3
· − · R
4. Rel i abi l i t y of HP compr essor f or 3000 hour s of oper at i on i s gi ven by:
9922 . 0 ) )
12000
3000
( exp( ) 3000 (
5 . 3
4
· − · R
5. Rel i abi l i t y of HP NGV f or 3000 hour s of oper at i on i s gi ven by:
9486 . 0 ) )
8000
3000
( exp( ) 3000 (
3
5
· − · R
6. Rel i abi l i t y of HP t ur bi ne f or 3000 hour s of oper at i on i s gi ven by:
9997 . 0 ) )
25000
3000
( exp( ) 3000 (
4
6
· − · R
7. Rel i abi l i t y of LP NGV f or 3000 hour s of oper at i on i s gi ven by:
8563 . 0 ) )
7000
3000
( exp( ) 3000 (
2 . 2
7
· − · R
8. Rel i abi l i t y of LP t ur bi n e f or 3000 hour s of oper at i on i s gi ven by:
9950 . 0 ) )
20000
3000
( exp( ) 3000 (
8 . 2
8
· − · R
9. Rel i abi l i t y of exhaust mi xer f or 3000 hour s of oper at i on i s gi ven by:
4. Syst ems Rel i abi l i t y 129
9243 . 0 ) )
7000
3000
( exp( ) 3000 (
3
9
· − · R
10. Rel i abi l i t y of ext er nal gear box f or 3000 hour s of oper at i on i s gi ven by:
9063 . 0 ) )
6500
3000
( exp( ) 3000 (
3
10
· − · R
11. Rel i abi l i t y of oi l t ank and f il t er f or 3000 hour s of oper at i on i s gi ven by:
8662 . 0 ) )
5000
3000
( exp( ) 3000 (
8 . 3
11
· − · R
Usi ng t he above val u es of i ndi vi dual r el i abi l i t i es, t he r el i abil i t y of t he syst em
i s gi ven by
4451 . 0 ) 3000 ( ) 3000 (
11
1
· ·
∏
· i
i S
R R
Fi gur e 4. 15 hazar d f unct i on f or t he engi ne.
3. Hazar d f unct i on of t he syst em.
Si nce al l t he i t ems of t h e syst em f ol l ow Wei bul l t i met of ai l ur e, t he hazar d
f unct i on i s gi ven by:
0
0.01
0.02
0.03
0.04
0.05
0.06
0
1
0
0
0
2
0
0
0
3
0
0
0
4
0
0
0
5
0
0
0
6
0
0
0
7
0
0
0
8
0
0
0
9
0
0
0
1
0
0
0
0
1
1
0
0
0
1
2
0
0
0
1
3
0
0
0
1
4
0
0
0
1
5
0
0
0
1
6
0
0
0
1
7
0
0
0
1
8
0
0
0
1
9
0
0
0
2
0
0
0
0
Time
H
a
z
a
r
d
f
u
n
c
t
i
o
n
4. Syst ems Rel i abi l i t y 130
1
) )( ( ) (
−
·
β
η η
β t
t h
The syst em hazar d f unct i on i s gi ven by:
∑
·
·
11
1
) ( ) (
i
i S
t h t h
It i s easy t o ver i f y t hat t he hazar d f unct i on of t h e syst em at t = 3000 and t =
7000 i s gi ven by:
000791 . 0 ) 3000 ( ·
S
h and 004796 . 0 ) 7000 ( ·
S
h
Fi gur e 4.15 depi ct s t h e hazar d f unct i on f or t he engi ne.
4. The expr essi on f or M TTF i s gi ven by:
)
1
1 (
β
η + Γ × · MTTF
By subst i t ut i ng t h e val ues of η and β, one can f i nd t he M TTF of di f f er en t
i t ems. Tabl e 4.9 gi ves t h e M TTF of di f f er ent i t ems.
Tabl e 4.9 M TTF of di f f er ent i t em of t he engi n e
It em M TTF (i n hour s)
LP com pr essor 13 395
LP st age 2 st at or 4 450
Int er medi at e casi ng 9 823
HP compr essor 10 800
HP NGV 7 144
HP t ur bi ne 22 650
4. Syst ems Rel i abi l i t y 131
LP NGV 6 202
LP t ur bi ne 17 800
Exhaust mi xer 6 251
Ext er nal gear box 5 804
Oi l t ank and fi l t er 4 525
Si nce t h e l owest M TTF i s 4 450 (LP st age 2 st at or ), t he M TTF of engi ne wi l l
be l ess t han 4 450.
5. M ean t i me t o f ai l ur e of a syst em subj ect t o pr event i ve mai nt enance i s
gi ven by:
) ( 1
) (
0
P S
T
S
pm
T R
dt t R
MTTF
P
−
·
∫
It i s gi ven t hat t he engi ne i s subj ect t o pr even t i ve mai nt enan ce ever y 1000
hour s of oper at i on. Thus, T
P
= 1000 ho ur s. Th e above expr essi on can be
eval uat ed usi ng numer i cal i nt egr at i on. The appr oxi mat e val ues of M TTF
p m
i s:
075 , 27
0369 . 0
06 . 999
) 1000 ( 1
) (
1000
0
≈ ≈
−
·
∫
S
S
pm
R
dt t R
MTTF
6. The mi ssi on r el i abi l i t y of t he engi n e i s gi ven by:
) (
) , (
) , (
b
m b
m b
t R
t t R
t t MR ·
wher e t
b
i s t he age of t he i t em at t he begi nni ng of t he mi ssi on and t
m
i s t he
mi ssi on dur at i on. Subst i t ut i ng t
b
= 5000 and t
m
= 1000, w e have
4. Syst ems Rel i abi l i t y 132
0548 . 0
02369 . 0
0013 . 0
) 5000 (
) 6000 (
) 5000 (
) 1000 5000 (
) , (
11
1
11
1
· · ·
+
·
∏
∏
·
·
i
i
i
i
m b
R
R
R
R
t t MR
7. The mai nt enance f r ee oper at i ng per i od sur vi vabi l i t y, M FOPS, f or t he
engi n e descr i b ed i s gi ven by:
∏
∏
·
·
× −
×
·
× −
×
·
11
1
11
1
) ] 1 ([
) (
) ] 1 ([
) (
) (
i
mf i
i
mf i
mf S
mf S
mf
t i R
t i R
t i R
t i R
t MFOPS
The abo ve equat i on can be eval uat ed f or t
mf
= 500 and f or i = 1, 2, ... et c.
Fi gur e 4.16 shows t he M FOPS val ues f or di f f er ent cycl es (not e t hat t h ese
val ues ar e der i ved wi t hout consi der i ng mai nt enan ce r eco ver y per i od M RP).
Fi gur e 4. 16 M FOPS val u e f or di f f er ent cycl es f or t he engi ne
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9
1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
1
9
Cycle number
M
F
O
P
S
5. M ai nt ai nabi l i t y and M ai nt enance 133
Chapter 5
Maintainability and Maintenance
M ai nt enance i s t he management of f ai l ur es and
t he assur ance of avai l abi l i t y
J Hessbur g
M ai nt ai nabi l i t y and mai nt enance has al w ays been i mpor t ant t o t he i ndust r y
as i t af f ect s t h e p er f or mance as w el l as t he f i nance. For commer ci al
ai r l i nes, mai nt enance cost s ar ound 10% of t h e ai r l i nes t ot al cost , as much as
f uel and t r avel agent s' commi ssi on (M Lam, 1995). Oper at or s/ user s woul d
l i ke t hei r syst em t o be avai l abl e and saf e t o oper at e when r equi r ed. One
shoul d be l ucky t o f i nd a smi l i ng cust omer w hen t h e syst em f ai l s and i t
t akes a l ong t i me t o r ecover t he f unct i onal i t y.
Ther e ar e sever al ways t hat desi gner s can pr o vi de maxi mum ut i l i t y of t hei r
pr oduct . One w ay i s t o bui l d i t ems/ syst ems t hat ar e ext r emel y r el i abl e (and
consequen t l y wi l l , al most cer t ai nl y, have a hi gher acqui si t i on cost ). Anot her
i s t o desi gn syst ems t hat ar e qui ck and easy t o r epai r w hen t hey f ai l .
Obvi ousl y, t he mai n obj ect i ve of t he desi gn er i s t o p r ovi de a r el i abl e and
saf e i t em at an af f or dabl e pr i ce.
M ai nt enance i s t he act i on necessar y t o sust ai n and r est or e t he
per f or mance, r el i abi l i t y and saf et y of t he i t em. The mai n obj ect i ve of
mai nt enance i s t o assur e t he avai l abi l i t y of t he syst em f or use w hen
r equi r ed. For ai r cr af t , mai nt enance f or ms an essent i al par t of
ai r wor t hi ness. The common o bj ect i ve of ai r cr af t mai nt enan ce, ci vi l or
mi l i t ar y, i s t o pr ovi de a f ul l y ser vi ceabl e ai r cr af t when i t i s r equi r ed by t he
oper at or at mi ni mum cost (Knot t s, 1996). However , mai nt enance cost s
money. Th e annual mai nt enance cost of pr odu ct i on asset s i n t h e Uni t ed
Ki ngdom i s est i mat ed i n excess of $13 bi l l i on, wi t h $2 bi l l i on wast ed
t hr ough i nef f i ci ent mai nt enance management pr act i ces (Knot t s, 1999).
5. M ai nt ai nabi l i t y and M ai nt enance 134
M ai nt enance al so account s f or appr oxi mat el y 10% of t he or gani sat i ons’
empl o yees and at l east 1015% of i t s oper at i ng cost s.
29. CONCEPT OF MAINTAINABILITY
In t he pr evi ous chapt er s, we showed t hat i t i s i mpor t ant f or t he
oper at or / user t o know t he r el i abi l i t y char act er i st i cs of t he i t em. We al so
r eco gni sed t hat i t i s al most i mpossi bl e f or any i t em t o mai nt ai n i t s f unct i on
f or ever , as f ai l ur e and t h e d egr adat i on of per f or mance i s i n evi t abl e. Thus,
f or t he user i t i s equal l y, or even mor e i mpor t ant t o know:
• When and how often maintenance tasks should be performed
• How they shoul d be performed
• How many people will be needed
• What skills they will need and how much training
• How much the restoration will cost
• How long the system wil l be down
• What facilities and equipment (special and general) will be required.
Al l t he above i nf or mat i on i s i mpor t an t as i t af f ect s t he avai l abi l i t y and t he
l i f e cycl e cost of t he syst em. One has t o appl y a sci ent i f i c di sci pli ne t o f i nd
answ er s t o t hese quest i ons.
M ai nt ai nabi l i t y i s t he sci ent i f i c di sci pl i ne t hat st udi es compl exi t y, f act or s
and r esour ces r el at ed t o t he mai nt enance t asks needed t o be per f or med by
t he user i n or der t o mai nt ai n t he f unct i onal i t y of a syst em, and w or ks out
met hods f or t hei r quant i f i cat i on, assessment , pr edi ct i on and i mpr ovement .
M ai nt ai nabi l i t y Engi neer i ng i s r api dl y gr owi ng i n i mpor t ance because i t
pr ovi des a ver y pow er f ul t ool t o engi neer s f or t h e quant i t at i ve descr i pt i on
of t he i nh er ent abi l i t y of t hei r syst em/ pr oduct t o b e r est or ed by p er f or mi ng
speci f i ed mai nt enance t asks. It al so cont r i but es t ow ar ds t he r edu ct i on of
mai nt enan ce cost s of a syst em dur i ng i t s ut i l i sat i on t o achi eve opt i mum l i f e
cycl e cost .
The mai nt ai nabi l i t y engi neer i ng f unct i on i nvol ves t he f or mul at i on of an
accept abl e combi n at i on of desi gn f eat ur es, w hi ch di r ect l y af f ect
mai nt enance and syst em suppor t r equi r ement s, r epai r pol i ci es, and
mai nt enance r esour ces. Some physi cal desi gn f eat ur es such as accessi bi l i t y,
vi si bil i t y, t est abi l i t y, compl exi t y and i nt er changeabi l i t y af f ect t h e sp eed and
ease wi t h whi ch mai nt enance can be per f or med.
5. M ai nt ai nabi l i t y and M ai nt enance 135
M ai nt ai nabi l i t y st udi es have t he f ol l owi ng obj ect i ves (R Knot t s 1996):
To guide and direct design decisions
To predict quantitative maintainability characteristics of a system
To identify changes to a system's design needed to meet operational
requirements
In t he t echni cal l i t er at ur e, sever al def i ni t i ons f or mai nt ai nabil i t y can be
f ound. For exampl e, t he US Depar t men t of Def ence' s M ILSTD721C (1966)
def i nes mai nt ai nabi l i t y as:
The measur e of t he abi l i t y of an i t em t o be r et ai ned i n or r est or ed t o
speci f i ed condi t i on w hen mai nt enance i s per f or med by per sonnel
havi ng speci f i ed ski l l l evel s, usi ng pr escr i bed pr ocedur es and
r esour ces, at each pr escr i bed l evel of mai nt enance and r epai r.
M ain t ain ab ili t y can b e expr essed i n t er ms o f mai n t en an ce f r equency
f act o r s, mai n t en an ce el ap sed t i mes and mai nt enance cost .
M ain t ain ab ili t y t her ef or e i s an i nheren t desi gn char act er i st i c deal i ng
w i t h t he ease, accu r acy, saf et y, and econo my i n t he p er f or mance o f
mai nt en ance f un ct i ons. M ai n t ai nabi l i t y req ui r ement s ar e d ef i ned in
concept ual d esi gn as p ar t of syst em oper at i onal r equi r ement s and
t he mai nt en an ce concep t . Anon (1992) descr i b es mai nt ai nab i l it y as:
The char act er i st i c of mat eri al desi gn and i nst all at i on t hat
det er mi nes t he r equi r ement s f or mai nt enance expendi t ur es
i ncl udi ng t i me, manpow er , per sonnel ski l l , t est equi pment ,
t echni cal dat a and f aci li t i es t o accompl i sh oper at ional obj ect i ves i n
t he user' s oper at i onal envi r onment .
One of t he common mi sper cept i on s i s t hat mai nt ai nabi l i t y i s si mpl y t he
abi l i t y t o r each a co mponent t o per f or m t he r equi r ed mai nt enance t ask
(accessi bi l i t y). Of cour se, accessi bi l i t y i s one of t he mai n concer ns f or many
mai nt enance engi n eer s. Fi gur e 5.1 i l l ust r at es an accessi bi l i t y pr obl em i n
one of t he ol der t wi nengi ne f i ght er ai r cr af t , Gl ost er Javel i n. Bef o r e an
en gi ne co uld be ch an ged, t h e j et pi pe h ad t o be di scon nect ed and
r emoved. To r emove t he j et pi p e i t w as necessar y f or a t echni ci an t o
gai n access t h ro ugh a hat ch and t hen b e su sp en ded upsi d e d own t o
r each t he cl amps and pi p es whi ch had t o b e di sconn ect ed. The job
5. M ai nt ai nabi l i t y and M ai nt enance 136
coul d on l y be achi eved b y t ou ch; t he i t ems wer e o ut si d e of t he
t ech ni ci an’ s f i el d of vi ew . The t echn i ci an h ad t o w or k hi s w ay down
bet w een t he engi ne an d t he ai r cr af t ’ s ski n, wi t h t ool s i n hi s h an d. For
saf et y r eason s, he was h el d by hi s ankl es, as shown i n f i gur e 5.1
(sou r ce: R Kn ot t s).
However , t her e ar e many ot her asp ect s t o be co nsi der ed ot her t han
accessi bi l i t y. M ai nt ai nabi l i t y shoul d al so consi der f act or s such as vi si bi l i t y,
t hat i s t he abi l i t y t o see a comp onent t hat r equi r es mai nt enance act i on,
t est abi l i t y (abi l i t y t o det ect syst em f aul t s and f aul t i sol at i on), si mpli cit y and
i nt er changeabi l i t y. Addi t i onal l y deci si onmaker s have t o be awar e of t he
envi r onment i n whi ch mai nt ai ner s oper at e. It i s much easi er t o mai nt ai n
an i t em on t he b ench, t han at t he ai r por t gat e, i n a war , amongst busy
mor ni ng t r af f i c, or i n any ot her r esul t or i ent ed and schedul edr i ven
envi r onment .
Fi gur e 5.1 Accessi bi l i t y concer n i n t he Javel i n f i ght er ai r cr af t
Anot her ar ea t o be consi der ed under mai nt ai nabi l i t y i s t r oubl eshoot i ng t he
var i ous modul es wi t hi n t he al l owed t i me, i . e. det er mi ni ng whet her t he
syst em i s saf e t o oper at e and, i f not , w hat act i on i s needed. For t he
commer ci al ai r l i nes, t her e i s usual l y l ess t han an hou r at t he gat e pr i or t o
t he ai r cr af t ’ s depar t ur e t o t he next dest i nat i on, w her eas f or a r aci ng car or
weapon syst em ever y second coul d be vi t al .
To meet t hese r equi r ement s, an easi l y manageabl e d evi ce i s needed w hi ch
can di agnose wi t h a hi gh degr ee of accur acy, whi ch modul es wi t hi n t he
syst em ar e at f aul t . It i s now wi del y accept ed t hat f al se r emoval s (of t en
r ef er r ed t o as No Faul t Found – NFF) cost about t he same as an act ual
f ai l ur e when t he comp onent under i nvest i gat i on i s r emoved and r epl aced.
Access Hatch
to Disconnect Jet Pipe
Items for
Disconnection
Access Hatch
to Disconnect Jet Pipe
Aircraft
Skin
5. M ai nt ai nabi l i t y and M ai nt enance 137
Reduci ng t he nu mb er of f al se r emoval s, t her ef or e, w oul d be a bi g cost
saver .
Devi ces w i t h t hese capabi l i t i es have been devel oped i n t he aer ospace,
For mul a 1 r aci ng car and l uxur y car i ndust r i es. For exampl e, t he Boei ng 777
i ncl u des an ' onboar d mai nt enance syst em' wi t h t h e obj ect i ve t o assi st t he
ai r l i nes t o avoi d exp ensi ve gat e d el ays and f l i ght cancel l at i ons. For si mi l ar
pur poses t he Fl i ght Cont r ol Di vi si on of t he Wr i ght Labor at or y i n t he USA has
devel oped a f aul t det ect i on/ i sol at i on syst em f or F16 ai r cr af t , whi ch all ow s
mai nt ai n er s, novi ce as wel l as exper t , t o f i nd f ai l ed compon ent s.
I n t he n ext sect i on, we di scuss t he mai nt ai nabi l i t y measur es and how t hese
measur es can be used f or ef f ect i ve mai nt enance management .
30. MEASURES OF MAINTAINABILITY
It i s ext r emel y i mpor t an t f or t he user t o have i nf or mat i on abou t t he
f unct i onal i t y, cost , saf et y, and ot her char act er i st i cs of t he pr oduct under
consi der at i on at t he begi nni ng of i t s op er at i ng l i f e. How ever , i t i s equal l y,
or even mor e i mpor t ant t o have i nf or mat i on about t h e char act er i st i cs wi t h
w hi ch t o def i ne t he mai nt enance t i me. M easur es of mai nt ai nabi li t y ar e
r el at ed t o t h e ease and econom y of mai nt enance such as; el apsed t i me t hat
an i t em spends i n t he st at e of f ai l ur e, manhour s r equi r ed compl et i ng a
mai nt enance t ask, f r equ ency of mai nt enan ce, and t he cost of mai nt enance.
As t he el apsed t i me has a si gni f i cant i nf l uence on t he avai l abi l i t y of t he
syst em, op er at or s w oul d l i ke t o know t he mai nt enance t i mes; not just t he
mean t i me but al so t he pr obabi l i t y t hat a mai n t enance t ask wi l l be
compl et ed wi t hi n a gi ven t i me. M ai nt enan ce el apsed t i mes ar e even
adver t i sed as a mar ket i ng st r at egy.
30.1 Maintenance ElapsedTime
The l en gt h of t h e el apsed t i me, r equi r ed f or t he r est or at i on of f unct i onal i t y,
cal l ed t i me t o r est or e, i s l ar gel y d et er mi ned at an ear l y st age of t he d esi gn
phase. The mai nt enance el apsed t i me i s i nf l uenced by t he co mpl exi t y of t he
mai nt enance t ask, accessi bi l i t y of t he i t ems, saf et y of t he r est or at i on,
t est abi l i t y, physi cal l ocat i on of t he i t em, as w el l as t he d eci si ons r el at ed t o
t he r equi r ement s f or t h e mai nt enan ce suppor t r esour ces (f aci l i t i es, spar es,
t ool s, t r ai ned p er sonnel , et c). It i s t her ef or e a f unct i on of t he
mai nt ai nabi l i t y and suppor t abi l i t y of t he syst em. It wi l l , of cour se, al so be
i nf l uenced by ot her f act or s dur i ng t he var i ous st ages of t he l i f e of t he
syst em but any bad deci si on made (ei t her expl i ci t l y or by def aul t ) dur i ng
5. M ai nt ai nabi l i t y and M ai nt enance 138
t he desi gn st age w i l l b e cost l y t o r ect i f y at a l at er st age and w i l l si gni f i cant ly
af f ect bot h t he oper at i onal cost s and syst em avai l abi l i t y.
1. Personnel factors which represent the influence of the skill, motivation,
experience, attitude, physical ability, selfdiscipline, training,
responsibility and other similar characteristics related to the personnel
involved;
2. Conditional factors which represent the influence of the operating
environment and the consequences of failure with the physical condition,
geometry, and shape of the item under restoration;
3. Environmental factors which represent the influence of factors such as
temperature, humidity, noise, lighting, vibration, time of the day, time of
the year, wind, noise, and others such as those similar to the maintenance
personnel factors during restoration.
Thi s mai nt ai nabi l i t y measur e can be r epr esent ed usi ng t he pr obabi l i t y t hat
t he mai nt enance t ask consi d er ed wi l l be compl et ed by a st at ed t i me. Si nce
t he mai nt enan ce el apsed t i me i s a r ando m var i abl e, one can use t he
cumul at i ve di st r i but i on f unct i on of t he el apsed t i me t o f i nd t h e per cen t age
of mai nt enance t asks t hat wi l l be compl et ed wi t hi n a speci f i ed t i me.
Mean Ti me t o Repai r
One appr oach f or measur i ng mai nt ai nabi l i t y i s t hr ough M ean Ti me t o
Repai r (M TTR). M TTR i s t he expect ed val ue of t he i t em' s r epai r t i me. Wi t h
t he knowl ed ge of t h e r el i abi l i t y and mai nt ai nabi l i t y of t he subsyst ems one
can eval uat e t he mai nt ai nabi l i t y of t he syst em, t hat i s, mean t i me t o r epai r
of t he syst em, M TTR
s
(Bi r ol i ni , 1994).
Assume t hat t he r el i abi l i t y bl ock di agr am of t he syst em has a ser i es
st r uct ur e wi t h n i t ems wi t h no r edundancy. Let M TTF
i
and M TTR
i
b e t he
mean t i me t o f ai l ur e and mean t i me t o r epai r of subsyst em i i n t he syst em.
Consi der an ar bi t r ar i l y l ar ge oper at i ng t i me T. Assumi ng t hat t he f ai l ur e
r at e of t he uni t i s const an t , t he expect ed numb er o f f ai l ur es of uni t i i n
dur i ng T i s gi ven by:
i
MTTF
T
(5.1)
The mean of t ot al r epai r t i me t o r epai r uni t i duri ng T i s gi ven by:
5. M ai nt ai nabi l i t y and M ai nt enance 139
i
i
MTTF
T
MTTR (5.2)
For t he whol e syst em, t he mean number of f ai l ur es i s gi ven by:
∑
·
n
i i
MTTF
T
1
(5.3)
For t he whol e syst em, t he mean of t ot al r epai r t i me i s gi ven by:
∑ ×
·
n
i i
i
MTTF
T
MTTR
1
(5.4)
Combi ni ng equat i on (5.3) and (5.4), we get t h e mean t i me t o r epai r at t he
syst em l evel , M TTR
s
, as:
∑
∑
·
·
·
n
i i
n
i i
i
s
MTTF
MTTF
MTTR
MTTR
1
1
1
(5.5)
Assumi ng const ant f ai l ur e r at e, t hat i s,
i
i
MTTF
1
· λ and
∑ ·
·
n
i
i s
1
λ λ , equation (5.5) can be written as:
∑ ·
·
n
i
i
s
i
s
MTTR MTTR
1
λ
λ
(5.6)
Exampl e 5.1
The M TTF and M TTR of f our subsyst ems i n a syst em ar e gi ven i n Tabl e 5.1.
Est i mat e t he syst em l evel mean t i me t o r epai r , M TTR
s
.
Subsyst em M TTF M TTR
1 200 24
2 500 36
5. M ai nt ai nabi l i t y and M ai nt enance 140
3 340 12
4 420 8
SOLUTION:
Appl yi ng equat i on (5.5), we get :
≈
+ + +
+ + +
·
420
1
340
1
500
1
200
1
420
8
340
12
500
36
200
24
s
MTTR 20 hours.
Mean Time to Repair MultiIndenture Case
Many complex systems are broken down into a number of levels of
indenture (LoI). For these systems, recovery of an LoI
i
unit is usually
achieved by the removal and replacement of LoI
i +1
items. In many cases, the
replacement LoI
i+1
item will not be the item that was removed. It may be a
new (i.e. unused) one or it may be one that was removed from another LoI
i
unit and subsequently recovered and put into stock for such an occasion.
Now, for such a system, the time to repair will be the time to remove and
refit the units at the next lower level of indenture. The elapsed time will
need to take into account logistic delays (i.e. waiting for equipment,
personnel, spares and any transport to and from the site at which the
maintenance work is to be done). This is discussed in more detail in Chapter
10.
Suppose a system is made up of n levels of indenture and a unit at LoI
i
is
made up of m
i
LoI
i +1
items. Suppose also that to recovery an LoI
i
unit, one
of the m
i
items is removed and replaced with average times, MTTRM
i,j
and
MTTRP
i ,j
respectively. Let us assume that the probability that item j is
rejected given that unit i has been removed is P
i j ,
then over an arbitrarily
long operating time T, the expected number of system failures is:
T
MTTF
1
where MTTF
1
is the mean time between failures of the system (over time
T).
Now, the probability that the failure was due to subsystem j is P
i,j
so the
mean time between failures due to subsystem j is
5. M ai nt ai nabi l i t y and M ai nt enance 141
MTTF
P
MTTF
P
j
j j j
1
1 1 1
1
1
1 1
,
, , ,
· · ·
λ λ
Assuming the system reliability block diagram is series and is series and
there are no redundancies.
The expected number of failures of subsystem j is
P
T
MTTF
T
MTTF
T
j
j
j 1
1 1
1 ,
,
,
· · λ
The expected time to recover the system given that subsystem j is the
cause of its failure is
MTTR MTTRM MTTRP
j j j 1 1 1 , , ,
· +
The exp ect ed t o t al t i me sp ent r ecover y t h e syst em due t o subsyst em j
f ai l ur es over t i me T i s t hen
P
MTTR
MTTF
T
MTTR
MTTF
T MTTR T
j
j j
j
j j 1
1
1
1
1
1 1 ,
, ,
,
, ,
· · λ
So, t he expect ed t ot al t i me spent r ecover i ng t he syst em b y subsyst em
exchan ge i s
P
MTTR
MTTF
T
MTTR
MTTF
T MTTR T
j
j
j
m
j
j j
m
j j
j
m
1
1
1 1
1
1 1
1 1
1
1 1 1
,
, ,
,
, ,
· · ·
∑ ∑ ∑
· · λ
Wher e m
1
i s t he number of subsyst ems. Th en t he mean t i me t o r ecover
t he syst em (by subsyst em exchan ge per syst em f ai l ur e) i s
MTTR MTTR
E
j
j
j
m
1
1
1
1
1
1
,
,
,
·
·
∑
λ
λ
To d et er mi n e t he t ot al mai nt enance t i me, w e w oul d have t o l o ok at t he
t i me spent r ecover i ng t he subsyst ems, b y subsubsyst em exchan ge and so
on dow n t o t he l ow est l evel compon ent s t hat ar e r ecover ed i n t hi s w ay and
5. M ai nt ai nabi l i t y and M ai nt enance 142
t hen add on any t i me spent r epai r i ng t he l ow est l evel compon ent s (par t s) i f
t hey can be r epai r ed but we wi l l l eave t hi s exer ci se unt i l our next book.
30.2 Maintenance Man Hour (MMH)
Al t hough el apsed t i me i s an ext r emel y i mpor t an t mai nt enance measur e,
one must al so consi der t he mai nt enance man hour s, M M H (al so know n as
mai nt enance l ab our ho ur s). The M M H i s an est i mat e of t he expect ed
“ spanner i nhand” t i m e and t akes i nt o account al l of t he mai nt enance t asks
and act i ons r equi r ed f or each syst em, subsyst em or compon ent r ecover y.
It shoul d be not ed t hat t h e M M H can be consi der abl y gr eat er t han t he
el apsed t i me as i t i s of t en possi bl e and so met i mes even necessar y t o
empl o y mor e t han one per son on any gi ven act i vi t y or t ask.
“ Wor k st udy” and “ t i me and mot i on” exer ci ses have gener at ed t abl es of
t i mes f or ever y concei vabl e mai nt enance act i on, f r om r el easi ng t he cat ches
t hat ar e used on access panel s t o i nspect i ng t he bl ades on a t ur bi ne usi ng a
bor oscope t o dr i l l i ng out a st ud t hat has shear ed af t er t oo much t or que has
been appl i ed t o i t , t o di sconn ect i ng and r econnect i ng al l of t he pi pes and
l eads when r emovi ng and r epl aci ng an engi ne.
In most cases, t hese t i mes ar e based on car r yi ng out t hese t asks and act i ons
i n i deal condi t i ons, i .e. i n a pr oper l y l i t w or kshop, w hi ch i s heat ed and
pr ovi des sh el t er f r om t he el ement s. They ar e gener al l y done when t he
compon en t s ar e i n pr i st i ne condi t i on f r ee f r om cont ami nat i on, cor r osi on or
damage. It i s al so gener al l y assumed t hat t he mechani c car r yi ng out each
act i on wi l l have been pr oper l y t r ai ned and f ami l i ar wi t h t he cor r ect
pr ocedur es. In pr act i ce, however , i t i s ver y r ar e f or al l of t hese i deal
condi t i ons t o be met so, t he act ual t i mes wi l l i nevi t abl y b e l onger t han
t hose used i n t he M M H pr edi ct i on.
M ai nt enance manhour s ar e usef ul i n t hei r own r i ght but ver y of t en t hey
ar e gi ven as a “ r at e” such as (M M H/ oper at i ng hour ), (M M H/ cycl e),
(M M H/ mont h), and (M M H/ mai nt enan ce t ask). For exampl e, el apsed t i mes
can be r educed (somet i mes) by i ncr easi ng t he number of peopl e i nvol ved i n
accompl i shi ng t he speci f i c t ask. How ever , t hi s may t ur n ou t t o be an
exp ensi ve t r adeof f , par t i cul ar l y when hi gh ski l l l evel s ar e r equi r ed t o
per f or m t he t asks. Al so, unl ess i t act ual l y r equi r es mor e t han one per son t o
do t he j ob, t her e i s l i kel y t o b e an “ i nt er f er ence f act or ” w hich means t hat
t he ef f i ci ency of each per son i s r educed. Ther ef or e, a pr oper bal ance
among el apsed t i me, l abour t i me, and per sonnel ski l l s at a mi ni mum
mai nt enan ce cost i s r equi r ed.
Commer ci al ai r l i nes and ai r f or ces use t he measur e M ai nt enance M anHour
per Fl i ght Hour (M M H / FH) as an i ndi cat or of t he mai nt ai nabi l i t y of t he
5. M ai nt ai nabi l i t y and M ai nt enance 143
ai r cr af t f or compar i son wi t h ot her si mi l ar ai r cr af t ei t her of an ol der
gen er at i on or mad e b y ano t her manuf act ur er . Thi s measur e may be used
t o deci de b et w een al t er nat i ves al t hough, i n many cases, i t wi l l be used t o
exer t pr essur e on t h e manuf act ur er t o make i mpr o vement s. The f ol l owi ng
expr essi on can be used t o eval uat e t he M M H/ FH:
hours flying Total
MNC MCMT t N MNC MPMT t N
FH MMH
cm pm
× × + × ×
·
) ( ) (
/
2 1
(5.7)
Wher e:
N
1
(t ) i s t he t ot al number of pr event i ve mai nt enance t asks dur i ng t
hour s, and N
2
(t ) i s t h e t ot al number of cor r ect i ve mai nt enance t asks. The
val ue t shoul d be equal t o t he oper at i onal l i f e of t he ai r cr af t .
M PM T = M ean pr event i ve mai nt enance t i me.
M CM T = M ean cor r ect i ve mai nt enan ce t i me.
M NC
p m
= M ean numb er of cr ew f or pr event i ve mai nt enance.
M NC
cm
= M ean nu mber of cr ew f or cor r ect i ve mai nt enance.
Not e t hat t hese est i mat ed mean val ues shoul d be w ei ght ed accor di ng t o
t he expect ed f r equency of each mai nt enan ce t ask as w e di d w hen
cal cul at i ng M TTR
s
above.
A pr obl em wi t h est i mat i ng t h e M M H/ FH met r i c i s t hat i t r el i es on t he
r el i abi l i t y of t he var i ous component s of t he syst em, whi ch may be age
r el at ed and wi l l , i nevi t abl y, depend on t he mai nt enance and suppor t
pol i ci es. For t hese r easons, t h e M M H/ FH may not r emai n const ant w i t h
ai r cr af t age. The i mpl i cat i on of usi ng such a met r i c i s t hat i t i s pr ef er ent i al
f or i t t o be mi ni mi sed, however , i t may act ual l y be bot h cheaper and yi el d a
hi gher l evel of avai l abi l i t y i f mor e t i me i s spent on mai nt enance, par t i cul ar l y
pr event at i ve mai nt enance.
30.3 Maintenance Frequency Factors
M ai nt ai nabi l i t y engi n eer i ng i s pr i mar i l y concer n ed wi t h desi gni ng a syst em
so t hat i t spends a mi ni mum t i m e i n mai nt enance, gi ven t hat i t needs
mai nt ai ni ng. Anot her char act er i st i c of syst em desi gn per t ai ni ng t o
mai nt ai nabi l i t y i s i n opt i mi si ng t he mi x b et ween pr even t at i ve and
cor r ect i ve mai nt enance.
The i deal syst em desi gn woul d al l ow t he oper at or s t o use t he syst em unt i l
j ust bef or e i t f ai l s but , w i t h enough not i ce of t h e i mpendi ng f ai l ur e so t hat
5. M ai nt ai nabi l i t y and M ai nt enance 144
t he oper at or can ch oose t o per f or m t he necessar y mai nt enan ce at t he most
oppor t une moment . In al l but a f ew cases, pr ognost i cs have, as yet , not
r each ed t hi s l evel of sophi st i cat i on. An al t er nat i ve appr oach i s bui l t i n
r edundancy and f aul t t ol er an t syst ems. These al l ow t he oper at or s t o def er
mai nt enance f or a l i mi t ed per i od or , i n cer t ai n ci r cumst an ces unt i l t he
backup syst em f ai l s.
Corrective maintenance can be expensive if the failure causes damage to
other parts of the system or if it stops the system from earning its keep.
However, redundant components will also add to the cost of the system and
may reduce its loadcarrying capacity. The spare wheel in cars takes up
space that could otherwise be used for carrying luggage, it also increases
the gross weight, which will reduce the performance of the car both by
reducing its rate of acceleration and increasing the fuel consumption.
It i s common pr act i ce f or mot or i st s t o r epl ace t yr es bef or e t he t r ead has
been co mpl et el y wor n away because i t i s unsaf e t o dr i ve on bal d t yr es. It i s
al so il l egal and t he penal t i es can b e bot h exp ensi ve and i nconveni ent . It i s
al so ver y easy t o i nspect t yr es f or wear so i t i s possi bl e t o l eave t hem unt i l
t he “ l ast mi nut e” or get t hem r epl aced when t he car i s not needed t hus
mi ni mi si ng t he i nconveni ence or l ack of avai l abil i t y.
Br ake pads ar e mor e di f f i cul t t o i nspect b y t h e owner . As a r esul t , many
car s ar e now f i t t ed wi t h pads t hat have an i nbui l t el ect r ode, whi ch causes a
war ni ng l i ght t o b e i l l umi nat ed on t he dash boar d when i t comes i nt o
cont act wi t h t he met al l i c di sc (due t o t he non conduct i ve par t of t he pad
bei ng wor n away). t hi s gener al l y gi ves t he dr i ver a suf f i ci ent war ni ng f or
hi m or her t o f i nd out w hat t he war ni ng l i ght means and t ake t he necessar y
cor r ect i ve act i on bef or e t he br akes beco me danger ous.
M ost mot or i st s have t hei r cam or t i mi ng bel t s r epl aced wi t hi n about 1000
mi l es of t he manuf act ur er ’ s r ecommended mi l eage possi bl y dur i ng a
r out i ne ser vi ce (sch edul ed mai n t enance) or at t he dr i ver / own er ’ s
conveni ence. In t hi s case, t h e owner has al most cer t ai nl y no way of
know i ng how much l onger t he b el t w i l l l ast and, i ndeed, i t i s l i kel y t o cost
t hem al most as much t o have t he b el t i nspect ed as i t w oul d t o have i t
r epl aced because of t h e amount of wor k i nvol ved. In t hi s case, t he ext ent
of t he damage t o t he en gi ne i f t he bel t br eaks i s l i kely t o cost a gr eat deal
mor e t han t hat of r epl aci ng t h e bel t ear l y. It woul d no doubt b e possi bl e t o
devi se a mo ni t or t hat coul d i ndi cat e wh en t he bel t was st ar t i ng t o wear but ,
whet h er i t woul d be pr act i cal i n t er ms of i t s si ze, r el i abi l i t y, cost and ext r a
wei ght i s ver y much open t o debat e.
Her e w e have seen f our di f f er ent sol ut i ons t o t he same pr obl em of avoi di ng
f ai l ur es and hence t he n eed f or cor r ect i ve mai nt enan ce. One of t he t asks
5. M ai nt ai nabi l i t y and M ai nt enance 145
of t he mai nt ai nabi l i t y engi neer i s t o det er mi n e whi ch, i f any of t hese, or
ot her si mi l ar appr oach es i s appr opr i at e t aki ng i nt o co nsi der at i on t h e cost s
and pr act i cal i t i es i n each ci r cumst ance.
Ther e i s cl ear l y a need t o st r i ke a bal ance. Pr event at i ve mai nt enance may
cause co mponent s t o be r epl aced unnecessar i l y (or at l east pr emat ur el y).
Al l owi ng a syst em t o r un unt i l i t f ai l s may maxi mi se t he t i mes bet ween
mai nt enan ce but f ai l ur es can b e expensi ve t o r ect i f y bot h because of t he
ext ent of t he damage caused and because of t h e l oss of avai l abi l i t y of t he
syst em whi l st i t i s bei ng mai nt ai ned. Pr o gnost i cs can hel p but t hese t oo
have t hei r own pr obl ems of r el i abi l i t y and t he need f or mai nt enan ce as wel l
as possi bl y addi ng t o t he wei ght , compl exi t y and cost of t he syst em.
30.4 Maintenance cost factors
For many syst ems/ pr odu ct s, mai nt enance cost s const i t ut e a maj or segmen t
of t he t ot al l i f ecycl e cost . Fur t her , exp er i en ce has i ndi cat ed t hat
mai nt enan ce cost s ar e si gni f i cant l y af f ect ed by desi gn deci si ons made
t hr oughout t he ear l y st ages of syst em devel opmen t . M ai n t ai nabi l i t y i s
di r ect l y concer n ed wi t h t he char act er i st i cs of syst em desi gn t hat wi l l
ul t i mat el y r esul t i n t he accompl i shment of mai nt enance at mi ni mum cost .
Thus, one way of measur i ng mai nt enance cost i s cost per mai nt enance t ask,
whi ch i s t he su m of al l cost s r el at ed t o el emen t s of l o gi st i cs suppor t whi ch
ar e r equi r ed t o p er f or m t he consi der ed mai nt enance t ask.
In addi t i on t o t he abo ve f act or s, t he f r equency wi t h whi ch each
mai nt enan ce act i on must be per f or med i s a maj or f act or i n bot h cor r ect i ve
and pr event i ve mai nt enance. Obvi ousl y t hi s i s gr eat l y i nf l uenced b y t he
r el i abi l i t y of t h e compon ent s but i t can al so be r el at ed t o t h e t yp e and
f r equency of t he mai nt enance per f or med. If a component i s r epai r ed t hen
i t i s l i kel y t hat t he t i me t o f ai l ur e f or t hat component wi l l be l ess t han i f i t
had been r epl aced by a new one. We wi l l r et ur n t o t he q uest i on of r epai r
ef f ect i ven ess i n Chapt er 6.
Per sonnel and human f act or consi der at i ons ar e al so of pr i me i mpor t ance.
These consi der at i ons i ncl ude t he exper i ence of t h e t echni ci an, t r ai ni ng, ski ll
l evel and number of t echni ci ans.
Suppor t consi d er at i ons co ver t he l ogi st i cs syst em and mai nt enance
or gani sat i on r equi r ed t o suppor t t he syst em. They i ncl ude t h e avai l abi l i t y
of spar e par t s, t echni cal dat a (manual s), t est equi pment and r equi r ed
speci al and gener al t ool s.
If a mai nt enance t ask r equi r es hi ghl y ski l l ed per sonnel , a cl ean envi r onment
equi pped w i t h expensi ve, speci al t ool s t h en i t i s unl i kel y, t hat i t w i l l pr ove
economi cal t o p er f or m t hi s t ask at f i r st l i ne or , possi bl y, even at second l i ne.
5. M ai nt ai nabi l i t y and M ai nt enance 146
How ever , i f t he mai nt ai nabi l i t y engi neer had d esi gned t he syst em so t hat
t hi s t ask coul d be don e b y per sonn el wi t h l ower ski l l l evel s usi ng st andar d
t ool i ng t hen i t mi ght have al l owed t he t ask t o be done i n t he f i el d wi t h a
possi b l e r educt i on i n t he t ur nar o und (or out of ser vi ce) t i me. I f , t he t ask i s
onl y l i kel y t o be done once i n t he syst em’ s l i f e dur i ng a maj or over haul
when i t woul d be at a cent r al mai nt enance uni t or r et ur n ed t o t he
manuf act ur er t hen such consi der at i ons may b e l ess r el evant . For exampl e,
t her e i s l i t t l e t o be gai n ed by maki ng i t easy t o r epl ace a br oken cam bel t by
t he si de of t h e r oad. The damage don e t o t he engi ne, as a r esul t of a f ail ed
cam bel t , wi l l mean t hat t h e engi ne wi l l ei t her have t o be r epl aced or
over haul ed/ r econdi t i oned bef or e i t i s l i kel y t o f unct i on agai n.
31. MAINTAINABILITY DEMONSTRATION
The obj ect i ve of t he mai nt ai nabi l i t y demonst r at i on i s t o show t hat t he
var i ous mai nt enance t asks can b e acco mpl i shed i n t he t i mes al l ot t ed t o
t hem. Gener al l y, t he most i mpor t ant i ssue i s whet her t h e syst em can be
r eco ver ed by subsyst em (or l i ne r epl aceabl e uni t – LRU) exchange wi t hi n
t he sp eci f i ed t i mes. It i s a common r equi r ement t hat each LRU can be
r emoved and r epl aced w i t hout i nt er f er i ng w i t h an y ot her LRU. Some of t he
ear l y j et f i ght er s w er e vi r t ual l y bui l t ar ound t he engi n e so t hat , i n or der t o
r epl ace t he engi ne, i t was not so much a quest i o n of r emovi ng t h e engi ne
f r om t he ai r cr af t as r emovi ng he ai r cr af t f r om t he engi ne.
A r ecen t i nnovat i on on commer ci al ai r cr af t i s t o use aut onomi cs, whi ch
si gnal ahead t o t he dest i nat i on any det ect ed f aul t s in t he mi ssi o n cr i t i cal
componen t s (i . e. t hose not on t he mi ni mu m equi pment l i st ). Thi s al l ows
t he mechani cs t o pr epar e t o r epl ace t h ese i t ems as so on as t he ai r cr af t has
r each ed t he gat e. I f such r epl acement s can be per f or med w i t hi n t he 50
mi n, or so, t ur nar ound t i me t hen i t wi l l not be necessar y t o f i nd a
r epl acement ai r cr af t or del ay t he depar t ur e. Anyon e w ho has seen t he f i l m
Bat t l e of Br i t ai n or Reach f or t he Sky w i l l r ecogni se t he i mpor t an ce of
t ur ni ng f i ght er ai r cr af t ar ound i n mi ni mum t i me when t he ai r f i el d may be
under at t ack f r om en emy bomber s and f ight er s. An ai r cr af t not i n t he ai r i s
bi t l i ke a duck ou t of w at er , i t i s par t i cul ar l y vul ner abl e and do ver y l i t t l e t o
def end i t sel f .
The demo nst r at i on i s al so expect ed t o gener at e r esul t s t hat can cont r i but e
t o t he whol e d evel opment pr ocess, i dent i f yi ng any r emai ni ng d ef i ci enci es
such as t h e desi gn of t he syst em and t he t est equi p men t , co mpi l at i on of
mai nt enance manual s, et c. Any mai nt ai nabi l i t y demonst r at i on woul d
i nvol ve t he f ol l ow i ng st eps:
5. M ai nt ai nabi l i t y and M ai nt enance 147
1. Ident i f y t he oper at i on and envi r onment al condi t i on i n whi ch t he
syst em i s l i kel y t o be used.
2. Si mul at e t he syst em f ai l ur es and per f or m cor r ect i ve mai nt enance
act i on. One shoul d al so r ecor d t he mai nt enance manhour s
r equi r ed t o compl et e t he r epai r t ask successf ul l y.
Fur t her , i t i s an i mpor t ant t o t ake car e of t h e f ol l ow ing i ssues dur i ng t he
demonst r at i on:
1. The t est must be on a sampl e of fi xed f i nal bui l d st andar d.
2. The t est condi t i ons must b e r epr esent at i ve, t he eq ui pment / t ool s,
mai n t enan ce manual s, l i ght i ng and si mi l ar f act or s must be car ef ul l y
consi der ed.
3. A mi x of r epai r er s r epr esent at i ve i n ski l l s, t r ai ni ng, and exper i ence of
t hose who woul d do t he act ual r epai r i n ser vi ce must conduct t he
r epai r .
Once we have t he r ecor ded r epai r t i me dat a f r om t he above pr ocedur e,
t hen i t i s easy t o ver i f y w het her t he mai nt ai nabi l i t y t ar get has been
achi eved usi ng t he f ol l owi ng pr ocedur e.
Let t
1
, t
2
, …, t
n
denot e t he obser ved r epai r t i mes t o compl et e t h e r epai r
t asks f or a sample of n uni t s. For n > 30, t he (1  α) 100 per cent conf i dence
l i mi t i s gi ven by:
n
s
z MTTR
α
+ (5.8)
Wher e z
α
i s t he z val ue (st andar d nor mal st at i st i c) t hat l ocat es an ar ea of α
t o i t s r i ght and can b e f ound f r om t he nor mal t abl e. For exampl e, f or a 95%
conf i dence l i mi t , t he z
α
i s gi ven by 1.645. M TTR and ' s' ar e gi ven by:
∑ ·
·
n
i
i
t
n
MTTR
1
1
, and
∑ −
−
·
·
n
i
i
MTTR t
n
s
1
2 2
) (
1
1
If t he t ar get mai nt ai nabi l i t y i s M TTR
*
, t hen t o demonst r at e t hat t he syst em
has achi eved t hi s, we have t o show t hat :
n
s
z MTTR MTTR
α
+ ≤
*
(5.9)
5. M ai nt ai nabi l i t y and M ai nt enance 148
Wh enever t he nu mber of r epai r t i me dat a i s l ess t han 30, we use t 
di st r i but i on; i n t hat case, t he condi t i on f or accept ance i s gi ven by:
n
s
t MTTR MTTR
n 1 ,
*
−
+ ≤
α
(5.10)
The val ue of t
α,n1
can b e obt ai n ed f r om t he t di st r i but i on t abl e show n gi ven
i n appendi x.
Exampl e 5.2
A mai nt ai nabi l i t y demonst r at i on t est i s car r i ed out on 20 par t s and t he
accompl i shed r epai r t i mes ar e show n i n Tabl e 5.2. If t he t ar get M TTR i s 20
hour s, check whet her t he syst em has achi eved t he t ar get mai n t ai nabi l i t y
usi ng 95% conf i dence l evel .
Tabl e 5.2. Recor d ed r epai r t i mes f or m a sampl e of 20 par t s i n hour s
8 6 12 20 24
12 9 17 4 40
32 26 30 19 10
10 14 32 26 18
SOLUTION:
Si nce t h e obser ved number of dat a, n i s l ess t han 30, we use t st at i st i c. The
M TTR and st andar d devi at i on, s, ar e gi ven by:
45 . 18
20
1
20
1
· ∑ ·
· i
i
t MTTR hours, ∑ − ·
·
n
i
i
MTTR t s
1
2
) (
19
1
= 10.06 hours
Fr om t he t dist r i but i on t abl e (see appendi x) w e get , t
α, n1
= 1. 729 (α = 0.05,
n1 = 19).
95% upper l i mi t f or M TTR i s gi ven by:
33 . 22
472 . 4
06 . 10
729 . 1 45 . 18 · × + · +
n
s
t MTTR
α
5. M ai nt ai nabi l i t y and M ai nt enance 149
Whi ch i s gr eat er t han 20 hour s, whi ch i s t he t ar get M TTR. Thus t he
achi eved M TTR i s si gni f i cant l y gr eat er t han t he r equi r ed M TTR and i s
t her ef or e not accept abl e.
32. MAINTENANCE
Accor di ng t o BS 4778, mai nt enance can be def i ned as:
The combi nat i on of al l t echni cal and admi ni st r at i ve act i ons, i ncl udi ng
super vi si on act i ons, i nt ended t o r et ain an i t em i n, or r est or e i t t o, a st at e i n
whi ch i t can per f or m a r equi r ed f unct i on.
In ot her wor ds, al l act i ons, whi ch keep t h e syst em r u nni ng and ensur e t hat
i t i s mai nt ai ned t o an accept abl e st andar d i n whi ch i t i s abl e t o oper at e at
t he r equi r ed l evel s ef f i ci ent l y and ef f ect i vel y. The obj ect i ves of
mai nt enan ce ar e t o:
1. Reduce the consequences of failure.
2. Extend the life of the system, by keeping the system in a proper condition
for a longer time. In other words, to increase the 'up¨ time of the system.
3. Ensure that the system is fit and safe to use.
4. Ensure that the condition of the system meets all authorised
requirements.
5. Maintain the value of the system.
6. Maintain reliability and achieve a high level of safety.
7. Maintain the system's availability and therefore minimise production and
quality losses.
8. Reduce overall maintenance costs and therefore minimise the life cycle
cost.
The pur pose of mai nt enance i s t o keep syst ems i n a st at e of f unct i oni ng i n
accor dance wi t h t hei r desi gn and t o r est or e t h em t o a si mi l ar st at e as and
when r equi r ed.
33. MAINTENANCE CONCEPT
The mai nt enance concept b egi ns wi t h a ser i es of st at ement s def i ni ng t he
i nput cr i t er i a t o whi ch t he syst em shoul d be desi gned. These st at emen t s
r el at e t o t h e mai nt enance t asks t hat shoul d be per f or med at each l evel of
5. M ai nt ai nabi l i t y and M ai nt enance 150
mai nt enan ce (or gani sat i onal , i nt er medi at e and depot ), t he t est equi p ment
and t ool s t hat shoul d be used i n mai nt ai ni ng t he syst em, t he ski l l l evel s of
t he mai nt enance per sonnel t hat per f or m t he i dent i f i ed t asks, mai nt enance
t i me const r ai nt s, and ant i ci pat ed mai nt enance envi r onment al r equi r ement s
(Knezevi c, 1997). A pr el i mi nar y mai nt enance concept i s devel oped dur i ng
t he co ncept ual desi gn st age, i s cont i nual l y updat ed, and i s a pr er equi sit e t o
syst em d esi gn an d d evel opment . M ai nt enance concep t at t h e desi gn phase
t ends t o ensur e t hat al l f unct i ons of desi gn and suppor t ar e i nt egr at ed wi t h
each ot her . The mai nt enance concept evol ved f r om t he def i ni t i on of
syst em oper at i onal r equi r emen t s del i neat es [ Bl anchar d et . al ., 1995]
• The ant i ci pat ed l evel of mai nt enance
• Over al l r epai r pol i ci es
• El ement s of mai nt enance r esour ces
• The or gani sat i onal r esponsi bi l i t i es f or mai nt enance
The mai nt enance concep t ser ves t he f ol l owi ng pur poses:
1. It provides the basis for the establishment of maintainability and
supportability requirements in the system design.
2. It provides the basis for the establishment of requirements for total
support which include maintenance tasks, task frequencies and time,
personnel quantities and skill levels, spare parts, facilities, and other
resources.
3. It provides a basis for detailing the maintenance plan and impacts upon
the elements of logistic support.
34. LEVELS OF MAINTENANCE
Compl ex syst ems can be consi der ed as mad e up of sever al l evel s of
i ndent ur e. A combat ai r cr af t t hat may be consi der ed as t he Level 0 (LoI0),
may be t hought of as consi st i ng f i ve subsyst ems: ai r f r ame, ar mament ,
avi oni cs, pr opul si on and gener al . The pr opul si on syst em t hen b eco mes a
LoI1 i t em t hat may consi st of t he engi n es, t he auxi l i ar y power uni t (APU)
and var i ous accessor i es i ncl udi ng cont r ol uni t s and p umps, each of w hi ch
may b e co nsi der ed as LoI2 i t ems. An engi ne i s t ypi cal l y an assembl y of a
number of modul es or LoI3 i t ems w hi ch, i n t ur n, may be mad e up of sub
assembl i es and par t s, LoI4 and 5 r espect i vel y.
At t he same t i me, t he mi l i t ar y t ypi cal l y di vi des i t s mai nt enance and sup por t
i nf r ast r uct ur e i nt o 3, 4 or 5 ech el ons, l i nes or [ mai nt enance] l evel s. “ Fi r st
5. M ai nt ai nabi l i t y and M ai nt enance 151
Li ne” , or “ OLevel ” i s f r om wher e t he syst ems ar e oper at ed. “ Second Li ne”
or “ ILevel ” i s t ypi cal l y t he mai n oper at i onal bases f r om whi ch t he
squadr ons ar e depl oyed. These ar e usual l y suppor t ed by a d epot or
mai nt enance uni t at “ Thi r d Li ne” or “ DLevel ” . Th e cont r act or , suppl i er or
or i gi nal equi pment manuf act ur er (OEM ) of t en pr ovi des a shadow f aci l i t y at
“ Four t h Li ne” ef f ect i vel y dupl i cat i ng t he Thi r d Li ne f aci l i t y’ s capabi l i t i es.
M ai nt enance l evel s ar e concer ned wi t h gr oupi ng t he t asks f or each
l ocat i on wher e mai nt enance act i vi t i es ar e per f or med. The cr i t er i a i n w hi ch
t he mai nt enance t asks sel ect ed at each l evel ar e; t ask compl exi t y,
per sonnel ski l l l evel r equi r ement s, speci al mai nt en ance equi pment and
r esour ces and economi c measur es. Wi t hi n t he scope of t he i dent i f i ed l evel
of mai nt enance, t he manuf act ur er and t he user shoul d def i ne a basi c r epai r
pol i cy t hat may var y f r om r epai r / r epl ace a par t ( LoI5, say) t o r epl ace t he
ent i r e syst em. The hi er ar chi es of achi evi ng mai nt enance t asks ar e di vi ded
i nt o t hr ee or f our l evel s.
34.1 User level (organisational)
Thi s t ype of mai nt enance l evel i s r el at ed t o al l mai nt enance t asks w hi ch ar e
per f or med on t he syst em whi l st i t i s on depl oyment or at i t s oper at i n g si t e.
Thi s woul d i ncl ude r epl eni shment t asks, e.g. r ef uel l i ng, r ear mi ng,
mai nt ai ni ng oi l l evel s, si mpl e condi t i on and per f or mance moni t or i ng
act i vi t i es, ext er nal adj ust ment s and r epl acement of l i ne r epl aceabl e uni t s
(LRU). Some mi nor r epai r s and r out i ne ser vi ci ng may al so come under t hi s
cat egor y.
34.2 Intermediate level
Int er medi at e mai nt enance l evel i s r el at ed t o al l mai n t enance t asks, w hi ch
ar e p er f or med at wor kshops (mobi l e, semi  mobi l e and/ or f i xed) wher e t he
syst ems woul d nor mal l y b e based. Common mai nt enance t asks
accompl i shed at t hi s l evel ar e det ai l ed condi t i on and per f or mance
moni t or i ng act i vi t i es, r epai r and r epl acement of maj or i t ems i n a syst em,
maj or over haul , syst em modi f i cat i on, et c. Per f or mi ng mai nt enan ce t asks at
t hi s l evel r equi r e hi gher per sonnel ski l l s t han t hose at or gani sat i onal l evel
and addi t i onal mai nt enance r esour ces. Tr adi t i onal l y, a r emoved LRU
woul d be r ecover ed, gener al l y by modul e (or shopr epl aceabl e uni t – SRU)
exchan ge, at t hi s l evel .
5. M ai nt ai nabi l i t y and M ai nt enance 152
34.3 Depot level
Depo t mai nt enance l evel i s r el at ed t o al l mai nt enance t asks, w hi ch ar e
accompl i shed beyond t he capabi l i t i es of i nt er medi at e l evel at r emot e si t es.
In t he UK syst em, “ Thi r d Li ne” r ef er s sp eci f i cal l y t o an oper at or own ed
f aci l i t y w her eas i n US nomencl at ur e “ DLevel ” al so i ncl udes
manuf act ur er / cont r act or f aci l i t i es. M ai nt enance t asks at t hi s l evel ar e
car r i ed out by hi ghl y ski l l ed speci al i st s at a speci al i sed r epai r f aci l i t y or t he
equi p ment pr oducer ’ s f aci l i t y. M ai nt enance t asks at depot l evel i ncl ude
compl et e o ver haul i ng and r ebui l di ng of t h e syst em, hi ghl y compl ex
mai nt enance act i ons, et c. Th ey woul d al so i ncl ude t asks whi ch may o nl y be
per f or med r ar el y, par t i cul ar l y i f t hey r equi r e expensi ve equi p men t or ar e
l i kel y t o t ake a l ong t i me.
34.4 HoleintheWall
Wi t h t he move t o ever gr eat er ef f i ci en cy and/ or mi ni mal cost s, t he
per cei ved n eed t o r educe manni ng l evel s and t he desi r e of OEM s t o
i ncr ease t hei r r evenu e b y ent er i ng t he “ af t er mar k et ” , t he “ hol ei nt he
wal l ” concept i s gai ni ng i n popul ar i t y. Thi s i s wher e t he onl y i nt r usi ve
mai nt enance t ask t he oper at or p er f or ms i s t o r emove t he LRU (at f i r st l i ne).
Thi s i s t hen passed t hr ou gh t hi s myt hi cal hol e i n t he wal l t o t he OEM or
mai nt enan ce cont r act or i n exchange f or a r epl acement (ser vi ceabl e) LRU.
The cont r act or t hen t akes t he LRU away t o a conveni ent l ocat i on wher e i t i s
r eco ver ed. Such cont r act s ar e of t en f und ed by f l eet hour ar r angement s
such as “ power byt hehour ” , see chapt er 12.
The advant age t o t h e oper at or s i s t hat t h ey can get on w i t h what t h ey ar e
i n busi ness f or ; put t i ng “ bums on seat s” or “ bombs on t ar get ” . It i s al so
ar gued, per haps mor e st r on gl y b y t he OEM t han t he oper at or , t hat havi ng
desi gn ed and bui l t t he LRU, t hey (t he OEM ) ar e t h e best peopl e t o t ake i t
apar t and r epai r i t . A secondar y advant age t o t he OEM , and agai n,
hopef ul l y t o t he oper at or , i s t hat because al l of t he m ai nt enance i s done i n
one pl ace, t he peopl e doi ng i t shoul d become mor e ef f i ci ent (as t hey see
t he sam e j ob mor e of t en) and t he i nser vi ce dat a (t i me t o f ai l ur e, cause of
f ai l ur e, i t ems r epai r ed or r epl aced, et c.) sho ul d be consi st ent and mor e
accur at e.
Bet t er dat a shoul d l ead t o i mpr oved f or ecast i ng, r educed l ogi st i c del ays,
mor e appr opr i at e mai nt enance pol i ci es and, ul t i mat el y, t o i mpr oved
desi gns.
5. M ai nt ai nabi l i t y and M ai nt enance 153
35. MAINTENANCE TASK CLASSIFICATION
Al l user s woul d li ke t hei r syst ems t o st ay i n a st at e of f unct i oni ng as l ong as
possi bl e or , at l east , as l ong as t hey ar e need ed. In or der t o achi eve t hi s, i t i s
necessar y t o mai nt ai n t he syst em’ s f unct i onal i t y dur i ng oper at i on, by
per f or mi ng appr opr i at e mai nt enance t asks. Thus, mai nt enance t ask can be
def i ned as a set of act i vi t i es t hat need t o be per f or med, i n a speci f i ed
manner , i n or der t o mai nt ai n t he f unct i onal i t y of t he i t em/ syst em.
Fi gur e 5.2 shows t he pr ocess of mai nt enance t ask, whi ch i s i ni t i at ed by t he
need f or mai nt enance d ue t o a r educt i on, or t er mi n at i on of t he
i t em/ syst em f unct i onal i t y. The execu t i on of a mai nt enance t ask r equi r es
r esour ces su ch as t he r i ght number and ski l l s of per sonnel , mat er i al ,
equi pment , et c. It al so r equi r es an appr opr i at e en vi r onment i n whi ch t he
mai nt enan ce act i vi t i es can be car r i ed out .
Fi gur e 5.2 Pr ocess of mai nt enance t ask
M ai nt enance t asks can be cl assi f i ed i nt o t he f ol l owi ng t hr ee cat egor i es:
1. corrective maintenance task
2. preventive (predictive) maintenance task
3. conditional maintenance task
Each maintenance task is briefly discussed in the following sections.
35.1 Corrective Maintenance Task
Cor r ect i ve mai nt enance t ask, CRT, i s a set of act i vi t i es, whi ch i s per f or med
w i t h t he i nt ent i on of r est or i ng t h e f unct i onal i t y of t he i t em or syst em, af t er
Maintenance task
activities
Need for
maintenance
Resources
Maintenance task
complete
Environment
Restore
functionality
5. M ai nt ai nabi l i t y and M ai nt enance 154
t he l oss of t he f u nct i onal i t y or p er f or mance (i . e. af t er f ai l ur e). Fi gur e 5.3
i l l ust r at es t ypi cal cor r ect i ve mai nt enan ce t ask act i vi t i es. The dur at i o n of
cor r ect i ve mai nt enance t ask, DM T
c
, r epr esent s t he el apsed t i me n eed ed f or
t he successf ul compl et i on of t he t ask. Cor r ect i ve mai nt enan ce t ask i s al so
r ef er r ed t o as an unschedul ed or unpl anned mai nt enance t ask.
Fi gur e 5.3 Act i vi t i es of t ypi cal cor r ect i ve mai nt enance t ask
35.2 Preventive Maintenance Task
Pr event i ve mai nt enance t ask, PM T, i s a mai nt enance act i vi t y t hat i s
per f or med i n or d er t o r educe t he pr obabi l i t y of f ai l ur e of an i t em/ syst em or
t o maxi mi se t he oper at i onal b enef it . Fi gur e 5.4 i l l ust r at es t he act i vi t i es of a
t ypi cal pr even t i ve mai nt enan ce t ask. The dur at i on of t h e pr event i ve
mai nt enan ce t ask, DMT
p
, r epr esent s t he el apsed t i me need ed f or t he
successf ul compl et i on of t he t ask.
MTBF
DMT
c
?
Item
Failed
Fault
Location
D
i
s
a
s
s
e
m
b
l
y
Repair or
Replacement
Assembly
T
e
s
t
a
n
d
C
h
e
c
k
Verification
Corrective maintenance
task activities
CMT start CMT complete
5. M ai nt ai nabi l i t y and M ai nt enance 155
Fi gur e 5.4 Act i vi t i es of a t ypi cal pr event i ve mai nt enance t ask
Pr event i ve mai nt enance t ask i s per f or med bef or e t he t r ansi t i on t o t he st at e
of f ail ur e occur s wi t h t he mai n obj ect i ve of r educi ng:
• The pr obabi l i t y of t he occur r ence of a f ai l ur e
• The consequences of f ai l ur e
Common pr event i ve mai nt enance t asks ar e r epl acement s, r en ewal and
over haul . These t asks ar e per f or m ed, at f i xed i nt er val s based on oper at i ng
t i me (e.g. hour s), di st ance (e.g. mi l es) or number of act i ons (e.g. l andi ngs),
r egar dl ess of t he act ual condi t i on of t he i t ems/ syst ems.
MTBF
DMT
p
Tp
Disassembly
R
e
p
l
a
c
e
m
e
n
t
Assembly
T
e
s
t
a
n
d
C
h
e
c
k
Verification
Preventive maintenance
task activities
PMT start
PMT complete
5. M ai nt ai nabi l i t y and M ai nt enance 156
35.3 Conditional (Predictive) Maintenance Task
Condi t i onal mai nt enan ce t ask, COT, r ecogni ses t hat a change i n condi t i on
and/ or per f or mance i s l i kel y t o pr ecede a f ai l ur e so t he mai nt enance t ask
shoul d be based on t h e act ual condi t i on of t he i t em/ syst em. COT does not
nor mal l y i nvol ve an i nt r usi on i nt o t he syst em and act ual pr event i ve act i on
i s t aken onl y w h en i t i s bel i eved t hat an i nci pi ent f ail ur e has been det ect ed.
Thus, t hr ough moni t or i n g of some condi t i on par amet er (s) i t woul d be
possi bl e t o i dent i f y t h e most sui t abl e i nst ant of t i me at whi ch pr event i ve
mai nt enan ce t asks shoul d t ake pl ace.
Fi gur e 5.5 Act i vi t i es of a t ypi cal condi t i onal mai nt enance t ask.
Fi gur e 5.5 i l l ust r at es t he act i vi t i es of a t ypi cal condi t i onal mai nt enance. The
dur at i on of condi t i onal mai nt enance t ask, DMT
m
, r epr esent s t he el apsed
t i me n eeded f or t he successf ul compl et i on of t he t ask.
In t he past , cor r ect i ve mai nt enance and pr event i ve mai nt enance t asks have
been popul ar among mai nt enance manager s. How ever , i n r ecent year s, t he
DMT
m
FMT
I
/ FMT
E
Inspection/
Examination
D
a
t
a
c
o
l
l
e
c
t
i
o
n
Condition
assessment
C
o
n
d
i
t
i
o
n
i
n
t
e
r
p
r
e
t
a
t
i
o
n
Decision
making
Conditional maintenance
task activities
COT start
COT complete
5. M ai nt ai nabi l i t y and M ai nt enance 157
di sadvant ages of t hese t asks have been r ecogni sed by man y mai nt enance
managemen t or gani sat i ons. The need f or t he pr ovi si on of saf et y, and
r educt i on of t he mai nt enance cost have l ed t o an i ncr easi ng i nt er est i n
usi ng condi t i onal mai nt enance t ask. Wai t i ng unt i l a componen t f ai l s may
maxi mi se t he l i f e obt ai ned f r om t hat co mponent but , i t s f ai l ur e may cause
si gni f i cant damage t o ot her par t s of t h e syst em and wi l l of t en occur at
i noppor t une t i mes causi ng a di sr upt i on t o t he oper at i on and i nconveni ence
t o t he user s. Rout i ne or schedul ed pr even t i ve mai nt enance, on t he o t her
hand, may be ver y conveni en t but i s l i kel y t o r esul t i n an i ncr ease i n t he
amount of mai nt enance needed b ecause par t s w i l l be r epl aced w hen t hey
have achi eved a f r act i on of t hei r expect ed l i f e.
36. MAINTENANCE POLICIES
The mai nt enance pol i cy def i nes whi ch t yp e of mai nt enance wi l l (nor mal l y)
be p er f or m ed o n t he var i ous component s o f t h e syst em. It i s d et er mi ned
by mai nt enance engi neer s, syst em pr oducer s and / or user s t o achi eve hi gh
saf et y, r el i abi l i t y and avai l abi l i t y at mi ni mu m cost . Wi t h r esp ect t o t he
r el at i on of t he i nst ant of occur r ence of f ai l ur e and t he i nst ant of per f or mi ng
t he mai nt enance t ask t he f ol l owi ng mai nt enance pol i ci es exi st :
1) Fai l ur eBased mai nt enance pol i cy, FBM , w her e cor r ect i ve mai nt enance
t asks ar e i ni t i at ed by t he o ccur r ence of f ai l ur e, i .e., loss of f unct i on or
per f or mance,
2) Ti meBased mai nt enance pol i cy, LBM , wher e pr even t i ve mai nt enance
t asks ar e p er f or med at pr edet er mi ned t i mes dur i ng oper at i on, at f i xed
l engt h of oper at i onal l i f e,
3) I nspect i onBased mai nt enance pol i cy, IBM , wher e condi t i onal
mai nt enance t asks i n t he f or m of i nspect i ons ar e per f or med at f i xed
i nt er val s of oper at i on, unt i l t he per f or man ce of a pr event i ve
mai nt enance t ask i s r equi r ed or unt i l a f ai l ur e occur s r equi r i ng
cor r ect i ve mai nt enance. Not e t hat t h e f ai l ur e coul d be due t o a
component of t he syst em t hat w as not bei ng subj ect ed t o I BM or i t
coul d have happen ed as a r esul t of some unpr edi ct ab l e ext er nal even t
such as f or ei gn obj ect damage or because t he i nspect i on i nt er val was
t oo l ong or t he i nspect i on w as i nef f ect i ve.
4) Exami nat i on Based mai nt enance pol i cy, EBM , wher e condi t i onal
mai nt enance t asks i n t he f or m of exami nat i ons ar e p er f or med i n
accor dance wi t h t h e moni t or ed condi t i on of t he i t em/ syst em, unt i l t he
execut i on of a pr even t i ve mai nt enance t ask i s needed or a f ai l ur e
occur s.
5. M ai nt ai nabi l i t y and M ai nt enance 158
The pr i nci pal di f f er ence b et ween t he above mai nt enance pol i ci es occur s at
t he t i me wh en t he mai nt enance t ask i s per f or med. The advant ages and
di sadvant ages of each mai nt enance pol i cy ar e br i ef l y descr i bed bel ow .
36.1 FailureBased Maintenance Policy
Fai l ur eBased mai nt enance pol i cy, FBM , r epr esent s an appr oach w her e
cor r ect i ve mai nt enan ce t asks ar e car r i ed out af t er a f ai l ur e has occur r ed, i n
or der t o r est or e t he f unct i onal i t y of t he i t em/ syst em consi der ed.
Conseq uent l y, t hi s appr oach t o mai nt enance i s known as br eakdown, post 
f ai l ur e, f i r e f i ght i ng, r eact i ve, or unschedul ed mai nt enance. Accor di ng t o
t hi s pol i cy, mai nt enance t asks of t en t ake pl ace i n ad hoc manner i n
r esponse t o br eakdown of an i t em f ol l owi ng a r epor t f r om t he syst em user .
A schemat i c pr esen t at i on of t he mai nt enan ce pr ocedur e f or t he f ai l ur e
based mai nt enance pol i cy i s pr esen t ed i n Fi gur e 5.6. Cor r ect i ve
mai nt enance t ask pr i or i t i es can r ange f r om " nor mal " , " ur gent " t o
" emer gency" . Th ese cat egor i es r ef l ect t h e nat ur e of t he r espo nse r at her
t han t he act ual act i ons done. Fai lur e based mai nt enan ce coul d be t he most
appl i cabl e and ef f ect i ve mai nt enance pol i cy i n si t uat i ons wher e:
Items for which the loss of functionality does not compromise the safety
of the user and/ or the environment or the failure has little or no
economic consequences (i.e. categories major and minor see
FMECA i n Chapter 11)
systems have buil tin redundancy or have been designed to be fault
tol erant
Fi gur e 5.6 Fai l ur eBased Mai nt enance Pol i cy
Advant ages of f ai l ure based mai nt enance
Item
Failed
Operating
time
5. M ai nt ai nabi l i t y and M ai nt enance 159
Impl ement at i on of FBM t o t h e abo ve si t uat i ons coul d l ead t o f ul l ut i l i sat i on
of t he op er at i ng li f e of t he i t em. Thi s means t hat t h e noncr i t i cal i t ems w i l l
have t he abi l i t y t o per f or m t hei r f unct i on(s) f or t he st at ed per i od of t i me
when t hey oper at e under st at ed condi t i ons. Thi s means t hat coef f i ci ent of
ut i l i sat i on, CU, whi ch is t he r at i o of t he M ean Dur at i on of Ut i li sed Li f e of t he
i t em (M DUL
F
) t o t h e exp ect ed oper at i ng l i f e (M TTF), of i t ems consi der ed
wi l l have val ue of 1. The user wi l l get maxi mum val ue out t h e component
w hen t he FB mai nt enance pol i cy i s appl i ed.
Di sadvant ages of f ai l ure based mai nt enance
Despi t e t h e advant ages of i mpl ement i n g t hi s pol i cy, i t has some
di sadvant ages when i t i s not cor r ect l y sel ect ed.
• The failure of an item will generally occur at an inconvenient
time.
• Maintenance acti viti es cannot be pl anned.
• It demands a lot of mai ntenance resources.
• The fai l ure of an i tem can cause a l arge amount of consequenti al
damage to other items in the system.
Anal ysi s of mai nt enance cost s have shown t hat a r epai r made af t er f ai l ur e
w i l l nor mal l y be t hr ee t o f our t i mes mor e expensi ve t han t he same
mai nt enan ce act i vi t y when i t i s wel l pl anned [ M obl ey (1990)] .
36.2 TimeBased Maintenance Policy
Some f ai l ur es can l ead t o econo mi cal consequences such as l oss of
pr oduct i on and t her ef or e a r educt i on i n pr of i t . Some f ai l ur es may have an
i mpact on t h e saf et y of t he user , passenger s, t hi r d par t i es and
envi r onment . Ther ef or e, i t i s desi r abl e t o pr event t hese f ai l ur es, i f possi bl e,
by car r yi ng out mai nt enance act i ons bef or e f ai l ur e occur s.
As t he mai n ai m i s t o r educe t h e pr obabi l i t y of occur r ence of f ai l ur e and
avoi d t he syst em br eakdow n, a t i mebased mai nt enance pol i cy i s
per f or med at f i x i nt er val s, whi ch i s a f unct i on of t he t i met of ai l ur e
di st r i but i on of t he i t em consi d er ed and i n so me cases i t may be adj ust ed by
t he syst em' s user . Thi s pol i cy i s ver y of t en cal l ed agebased, l i f ebased,
pl anned or schedul ed mai nt enan ce. The r eason f or t hat i s t he f act t hat t he
5. M ai nt ai nabi l i t y and M ai nt enance 160
mai nt enan ce t ask i s p er f or m ed at a pr edet er mi n ed f r equency, whi ch may
be based on, f or exampl e, oper at i ng t i mes such as, hour s, year s, mi l es,
number of act i ons or any ot her uni t s of use, t hat make i t i s possi bl e t o pl an
al l t asks and f ul l y suppor t t hem i n advance. A schemat i c pr esent at i on of
t i mebased mai nt enance pr ocedur e i s pr esent ed i n Fi gur e 5.7. The
f r equency of mai nt enan ce t ask, FMT
L
, is det er mi ned even bef or e t he
i t em has st ar t ed f unct i oni ng. Thus, at t he pr edet er mi ned l engt h of
oper at i onal l i f e speci f i ed, pr event i ve mai nt enan ce t asks t ake pl ace. The
t i mebased mai nt enance pol i cy coul d be ef f ect i vel y appl i ed t o
i t ems/ syst ems t hat meet som e of t he f ol l owi ng r equi r ement s:
1. the probability of occurrence of failure is reduced
2. the likely consequences of failure is 'catastrophic¨ (e.g. loss of life or
serious injury)
3. the total costs of applying this policy are substantially lower than the
alternatives
4. the condition of the system, or its consisting items, cannot be monitored
or is impractical or uneconomical.
Advant ages of t i mebased mai nt enance pol i cy
One of t he mai n advant ages of t hi s mai nt enance p ol i cy i s t he f act t hat
pr event i ve mai nt enance t asks ar e p er f or med at a pr edet er mi ned i nst ant of
t i me wh en al l mai nt enance suppor t r esour ces coul d be pl anned and
pr ovi ded i n ad vance, and p ot ent i al cost l y o ut ages avoi ded. For f ai l ur es,
whi ch coul d have cat ast r ophi c consequ ences t o t he user / oper at or and
envi r onment (Cher nob yl , Bhopal , Pi per Al pha and si mi l ar ) i t may be t he
onl y f easi bl e opt i on. Ti mebased mai nt enance has many advan t ages over
f ai l ur ebased mai nt enan ce, whi ch ar e summar i sed i n t he f ol l o wi ng l i st :
1. Maintenance can be planned ahead and performed when it is convenient
from the operational and logistics point of view.
2. The cost of lost production and of consequential damage can be reduced.
3. Downtime, the time that the system is out of service, can be minimised.
4. Safety can be improved.
5. M ai nt ai nabi l i t y and M ai nt enance 161
Fi gur e 5.7 Ti me Based M ai n t enan ce Pol i cy
Di sadvant ages of t i me based mai nt enance pol i cy
In spi t e of t he advant ages gi ven above, t he t i mebased mai nt enan ce p ol i cy
has sever al di sadvan t ages t hat must be mi ni mi sed. Thi s pol i cy coul d be
uneconomi cal because t he maj or i t y of i t ems ar e p r emat ur el y r epl aced,
i r r espect i ve of t hei r condi t i on. In man y i ndust r i es t hi s pol i cy i s now onl y
used under sp eci al condi t i ons because i t i s ver y cost l y, and al so because i t s
ef f i ci ency i n r educi ng f ai l ur es i s not al w ays suppor t ed by exper i ence. A
summar y of t h e di sad vant ages of t i mebased mai nt enance pol i cy i s l i st ed
bel ow .
1. Timebased maintenance is performed irrespective of the condition of the
system. Consequently, a large number of unnecessary tasks will be
carried out on a system that could have been operated safely for a much
longer time.
2. The tasks may require higher numbers of skilled mechanics.
3. If the time to perform the maintenance is greater than the time the system
would normally be idle (eg overnight) then because of the frequency, it
could cause higher levels of unavilability.
4. It cannot guarantee the elimination of all failures and will do nothing to
reduce nonagerelated failures.
5. Increasing the frequency of maintenance tasks may lead to an increase in
the probability of human errors in the form of maintenanceinduced
failures.
6. Reducing the probability of failure by prematurely replacing components
means that the coefficient of utilisation of the item/system, CU
L
, will
have a value much less than one.
Preventive
Task
System in use
Predetermined
Time Tp
Operating time
FMT
L
5. M ai nt ai nabi l i t y and M ai nt enance 162
36.3 Condition Based Maintenance (Predictive
Maintenance)
The need f or t h e pr ovi si on of saf et y, i ncr eased syst em avai l abi l i t y, and
r educed mai nt enance cost s have l ed t o an i ncr easi ng i nt er est i n
devel opment of al t er nat i ve mai nt enance p ol i ci es. A pol i cy w hi ch
over comes man y of t h e di sadvant ages of t he pr evi ous mai nt enance pol i ci es
(f ai l ur ebased and t i mebased), and has pr oved i t s abi l i t y t o ext end t he
oper at i ng l i f e of a syst em w i t hout i ncr easi ng t he r i sk of f ail ur e is condi t i on
based mai nt enance, CBM . CBM i s al so know n as pr edi ct i ve mai nt enan ce.
Condi t i onbased mai nt enance can be d ef i ned as: " M ai nt enance car r i ed out
i n r esponse t o a si gni f i cant det er i or at i on i n a uni t as i ndi cat ed by a change
i n t he moni t or ed par amet er s of t he uni t ' s condi t i on or per f or mance" [ Kel l y
& Har r i s (1978)] . Thi s means t hat t he pr i nci pl e r eason f or car r yi ng out
mai nt enan ce act i vi t i es i s t he change or d et er i or at i o n i n condi t i on and/ or
per f or mance, and t he t i me t o per f or m mai nt enan ce act i ons i s det er mi ned
by moni t or i ng t h e act ual st at e of t he syst em, i t s per f or mance and/ or ot her
condi t i on par amet er s. Thi s shoul d mean t he syst em i s oper at ed i n i t s most
ef f i ci ent st at e and t hat mai nt enance i s onl y per f or med when i t i s cost 
ef f ect i ve. A sch emat i c pr esen t at i on of condi t i onbased mai nt enance
pr ocedur e i s pr esent ed i n Fi gur e 5.8. Thi s pol i cy i s wor t h appl yi ng i n
si t uat i ons w her e:
1 The st at e of t he syst em i s descr i bed b y one or mor e condi t i on
par amet er s.
2 The cost of t h e condi t i on moni t or i ng t echni qu e i s l ow er t han t he
expect ed r educt i on i n over al l mai nt enance cost s.
3 Ther e i s a hi gh pr obabi l i t y of det ect i ng pot en t i al l y cat ast r ophi c f ai l ur es
(bef or e t hey happen).
Fi gur e 5.8 Condi t i on based mai nt enance pol i cy
Unsatisfactory
Preventive
Task
System in use
Inspection time
Operating time
Satisfactory
C
o
n
d
i
t
i
o
n
System in use
5. M ai nt ai nabi l i t y and M ai nt enance 163
The condi t i onbased mai nt enan ce i s a condi t i o n or per f or mancedr i ven
pr event i ve mai nt enan ce. Thi s means t hat t he t i mi n g of t he mai nt enance
t ask i s not si mpl y a f unct i on of t he mean t i met of ai l ur e. The pr i nci pl e of
condi t i onbased mai nt enan ce t her ef or e i s based on t he way of moni t or i ng
t he condi t i on par amet er s of a syst em gi vi ng t hr ee di f f er ent t ypes of
condi t i on moni t or i n g:
1 Inspect i on
Inspect i on i s gener al l y per f or med at r egul ar i nt er val s usi ng any of a number
of nondest r uct i ve t est (NDT) pr ocedur es whi ch ar e desi gned t o det er mi ne
w het her t he condi t i on of t he (i nspect ed) i t em i s sat i sf act or y or
unsat i sf act or y and hence whet her f ur t her act i on i s r equi r ed.
2 Exami nat i on
Thi s i s a condi t i onmoni t or i ng t ask, whi ch pr esent s a numer i cal descr i pt i on
of t h e condi t i on of t he i t em at t hat moment t hr ou gh r el evant condi t i on
pr edi ct or s. The r esul t s di r ect l y af f ect t he sch edul i ng of t he next
exami nat i on. Thi s i s possi bl e because of t he uni que pr op er t i es and
char act er i st i cs of t he r el evant condi t i on pr edi ct or .
3  Perf ormance Trend Moni t ori ng
For propulsion or energy producing systems, in particular, the
performance may be expressed as a ratio of the output to input, e.g. miles
per gallon, kilometres per litre, thrust per kilogram or (mega)watts per
tonne. As the system deteriorates, usually through wear but also through
damage, these ratios may show signs of decreasing. For systems operating
in relatively constant conditions (e.g. constant ambient temperature,
pressure and output), consistent changes in the specific fuel consumption
(SFC) will almost certainly be indicative of a deterioration in the system
which will need some form of maintenance to restore it to an acceptable
level. For systems that are operated in an inconsistent manor for which the
environmental conditions may be in a constant state of change, the SFC may
be subject to considerable noise and hence any deterioration will only be
apparent by using sophisticated trending algorithms, such as Kalman
Filtering.
5. M ai nt ai nabi l i t y and M ai nt enance 164
36.3.1Setting up conditionbased maintenance policy
In or der t o i mpl ement CBM pol i cy, i t i s necessar y t o use t he f ol l owi ng
managemen t st eps t hat ar e shown i n Fi gur e 5.9
Ident i f i cat i on and sel ect i on of mai nt enance si gni f i cant i t ems
The f i r st r equi r emen t of i mpl emen t i ng CBM i s t o d eci de whi ch i t ems of t he
syst em shoul d be moni t or ed, si nce i t i s l i kel y t o be bo t h unecono mi cal and
i mpr act i cal t o moni t or t hem al l . Ther ef or e, t he f i r st st ep of t he condi t i on
based mai nt enance deci si on pr ocess i s a compr ehensi ve r evi ew of al l i t ems
i n a syst em, i n or der t o i dent i f y t he mai nt enan ce si gni f i cant i t ems, M SI s.
These ar e i t ems whose f ai l ur es coul d b e saf et y cr i t i cal , envi r onment al l y
damagi ng or r evenue sensi t i ve. Thus, each i t em wi t h i n t he syst em shoul d
be anal ysed f r om t he poi nt of vi ew of f ai l ur e, especi al l y t he consequences
of f ai l ur e. The most f r equ ent l y used engi n eer i ng t ool s f or per f or mi ng t hi s
t ask i s a Fai l ur e M ode Ef f ect and Cr i t i cal i t y Anal ysi s, FM ECA and Rel i abi l i t y
Cent r ed M ai nt enance, RCM (see al so Chapt er s 6 and 11). Car e sh oul d be
t aken t o ensur e t hat al l of t he mai nt enan ce si gni f i cant i t ems ar e i dent i f i ed
and li st ed.
Ident i f i cat i on and sel ect i on of condi t i on paramet ers
Once t he mai nt enan ce si gni f i cant i t ems ar e i dent i f i ed i t i s n ecessar y t o
det er mi ne al l moni t or abl e par amet er s whi ch descr i be t hei r condi t i on or
per f or mance. The condi t i on par amet er can be def i ned as a measur abl e
var i abl e abl e t o di spl ay di r ect l y or r ef l ect i ndi r ect l y i nf or mat i on about t he
condi t i on of an i t em at any i nst ance of oper at i ng t i me. Ideal l y, mai nt enance
engi n eer s woul d l i ke t o f i nd many condi t i on/ par amet er s whi ch can be
moni t or ed and w hi ch accur at el y r ef l ect s t he condi t i on / per f or man ce of t he
syst em. In pr act i ce t h er e ar e t wo di st i ngui shabl e t ypes of condi t i on
par amet er s w hi ch ar e abl e t o achi eve t hi s (Knezevi c et al , 1995):
A. Rel evant Condi t i on Indi cat or, RCI
The Rel evan t Condi t i on Indi ct or , RCI, i s a par amet er t hat d escr i bes t he
condi t i o n of an i t em dur i ng i t s oper at i n g t i me and i t i ndi cat es t he condi t i on
5. M ai nt ai nabi l i t y and M ai nt enance 165
of t he i t em at t he i nst ant of i nspect i on. The numer i cal val ue of RCI
r epr esen t s t he l ocal val ue of t he condi t i on of an i t em/ syst em at t h e t i me of
i nspect i on. Thi s t ype of condi t i on par amet er i s usual l y r el at ed t o t he
per f or mance. However , RCI i s not abl e t o pr edi ct t he f ut ur e devel opment of
t he condi t i on of t he consi der ed i t em/ syst em. Typi cal exampl es of t he RCI
ar e p er f or mance, t he l evel of vi br at i on, l evel of oi l , pr essur e, t emper at ur e,
et c. It i s necessar y t o st r ess t hat t h e RCI coul d have an i dent i cal val ue at
di f f er en t i nst ances of oper at i ng t i me.
Fi gur e 5.9 Fl ow of condi t i on based mai nt enance
Rel evant Condi t i on Predi ct or (RCP)
The Rel evant Condi t i on Pr edi ct or , RCP, i s a par amet er , whi ch descr i bes t he
condi t i on of an i t em at ever y i nst ant of oper at i ng t i me. Usual l y t hi s
par amet er i s di r ect l y r el at ed t o t he shap e, geomet r y, wei ght , and ot her
char act er i st i cs, whi ch descr i be t he condi t i on o f t he i t em under
consi der at i on. Th e RCP r epr esent s t he condi t i on of t he i t em/ syst em whi ch
i s most li kel y t o be af f ect ed b y a gr adual det er i or at i on f ail ur e such as w ear ,
cor r osi on f at i gue cr ack gr owt h. The gener al pr i nci pl es of t he RCP ar e
di scussed by Knezevi c (1987). Typi cal exampl es of RCP ar e: t hi ckness of an
System/ items selection
Condition parameters
selection
Condition monitoring
techniques selection
Collecting data &
information
fault diagnosis
Condition
assessment &
Conditionbased
maintenance task
accepted
unaccepted
5. M ai nt ai nabi l i t y and M ai nt enance 166
i t em, cr ack l engt h, dept h of t yr e t r eads, et c. The RCP cannot have i dent i cal
val ues at t wo or mor e i nst ance of t i me. The nu mer i cal val ue of t he r el evant
condi t i on pr edi ct or at any i nst ant of oper at i ng t i me quant i f i es t he
cumul at i ve val ue of t he condi t i on of an i t em/ syst em at t h e t i me of
exami nat i on.
Sel ect i on of condi t i on moni t ori ng t echni que
Havi ng i dent i f i ed t he mai nt enance si gni f i cant i t em and t he associ at ed
condi t i on par amet er (s), t he n ext st ep i s t o sel ect t he sui t abl e moni t or i ng
t echni que, whi ch wi l l be used t o i nspect and exami ne each condi t i on
par amet er .
The condi t i on moni t or i ng t echni que i s a devi ce used t o i nspect or exami ne
an i t em i n or der t o pr o vi de dat a and i nf or mat i on abo ut i t s condi t i on at any
i nst ance of oper at i ng t i me. Numer ous condi t i on moni t or i ng t echni qu es, f or
i nst ance, NDT t echni ques, p er f or man ce, vi br at i on, et c ar e avai l abl e f or use
by mai nt enan ce engi neer s i n or der t o d et er mi ne measur abl e val ue of
condi t i o n par amet er . It i s i mpor t ant t o under st and t he b ehavi our of t he
f ai l ur e t hat t he i t em exhi bi t s so t hat t he most ef f ect i ve mo ni t or i ng
t echni ques can be chosen.
The d eci si on as t o w hi ch condi t i onmoni t or i ng t echni ques ar e sel ect ed
depends gr eat l y on t he t yp e of syst em, t he t ype of condi t i on par amet er
and, i n t he end, on cost and saf et y. Once t he d eci si on i s made as t o whi ch
t echni ques ar e t o b e used, i t i s p ossi bl e t o def i ne t he equi pment or
i nst r ument t hat wi l l be needed t o car r y out condi t i on moni t or i ng.
Col l ect i ng dat a and i nf ormat i on
The phi l osophy of condi t i on moni t or i n g i s t o assess t he condi t i on of an
i t em/ syst em by t he use of t echni ques whi ch can r ange f r om human sensi ng
t o sophi st i cat ed i nst r ument at i on, i n or der t o det er mi ne t he n eed f or
per f or mi ng pr event i ve mai nt enance t asks. Wi t h t he i ncr eased i nt er est i n
condi t i on moni t or i ng i n r ecent year s t her e have been a number of
devel op ment s i n t he t echni ques t hat ar e used t o col l ect dat a and pr o vi de
i nf or mat i on, w hi ch hel ps mai nt enan ce engin eer s assessi ng t he condi t i on of
an i t em or a syst em. These devel opmen t s have mad e i t possi bl e t o obt ai n
mor e r el i abl e i nf or mat i on on t he condi t i on of t he syst em. In many
i nst ances such i nf or mat i on i s used t o i nsur e t hat t he st at us of t h e syst em
wi l l cont i nue t o be i n a f unct i oni ng st at e wi t ho ut si gni f i cant r i sk of
5. M ai nt ai nabi l i t y and M ai nt enance 167
br eakdown, and i n some i nst ances t o make a deci si on on t he t i mi n g of
when mai nt enan ce t asks shoul d be per f or med. The met hod of dat a
col l ect i on can be cl assi f i ed i nt o t he f ol l owi ng cat egor i es:
Onl i ne dat a col l ect i on and moni t ori ng
Onl i ne dat a col l ect i on and moni t or i ng uses i nst r ument at i on f i t t ed t o t he
syst em w hi ch t ak es cont i nuous measur ement s of t h e condi t i on par amet er s.
These may t hen be anal ysed b y an onboar d comput er t o det er mi ne
whet her t her e has been a change i n t he condi t i on of t he i t em/ syst em and
whet h er t hat chan ge r equi r es an y act i on. The benef i t of usi ng onl i ne
moni t or i n g i s t o r educe t he n eed f or human i nt er vent i on and mi ni mi se t he
pr obabi l i t y of a f ai l ur e occur r i ng bet w een i nspect i ons.
Of f l i ne col l ect i on and moni t ori ng
Of f l i ne col l ect i on and moni t or i ng i s per i odi c measur emen t of a condi t i on
of an i t em/ syst em or con t i nuous dat a col l ect i on whi ch i s anal ysed
r emot el y. Thi s t ype of met hod i nvol ves ei t her t he col l ect i on of dat a usi ng a
por t abl e dat a col l ect or , or t aki ng a physi cal sampl e, f or exampl e, l ubr i cat i on
oi l sampl es f or anal ysi s of cont ami nat i on and debr i s cont ent . Per i odi c
moni t or i ng t h er ef or e pr ovi des a way of det ect i ng pr ogr essi ve f aul t s i n a
way t hat may be cheaper t han t he onl i ne syst em.
Fi gur e 5.10. Condi t i on moni t or i ng and condi t i on assessmen t
Condi t i on assessment
5. M ai nt ai nabi l i t y and M ai nt enance 168
The assessment of t he condi t i on of an i t em/ syst em (Fi gur e 5.10) can r ange
f r om human exper i ence t o sophi st i cat ed i nst r ument at i on. Th e l ast f ew
decades have seen a number of devel opment s i n t he met hods w hi ch ar e
used t o hel p t he mai nt enance engi neer s assess and di agnose t he condi t i on
of an i t em/ syst em and pr ovi de t hem wi t h i nf or mat i on on whi ch t o base
t hei r deci si on. Once condi t i on moni t or i ng sensor s have been i nst al l ed and
dat a ar e bei ng col l ect ed, i t i s necessar y t o have r el i abl e met hods of
i nt er pr et i ng t he dat a t o i den t i f y whet her t he consi der ed i t em i s under goi ng
a t r ansi t i on f r om t he nor mal t o abnor mal condi t i on and i n many cases t o
i dent i f y t he causes of t he chan ges.
Ef f ect i ve condi t i onbased mai nt enance r equi r es a l ar ge number of
measur ement s t aken cont i nuousl y or at i nt er val s t hat assur e r ecogni t i on of
change i n t he condi t i on of t he i t em/ syst em i n suf f i ci ent t i me t o avoi d t he
need f or any cor r ect i ve act i on. The vol ume of dat a necessar y t o accur at el y
det er mi ne t he condi t i on of t he i t em/ syst em can r equi r e an excessi ve
amount of t i me t o pr o cess and anal yse. Consequent l y, t he demand t o
mani pul at e and pr ocess l ar ge amount s o f dat a ver y qui ckl y has l ead t o t he
devel opment of t ool s such as Ar t i f i ci al I nt el l i gence, AI, t o assi st engi neer s t o
gai n maxi mum val ue f r om t h e dat a.
In r ecent year s, Ar t i f i ci al Int el l i gence t echni ques such as Exper t Syst em,
Neur al Net wor ks and Fuzzy Logi c have been appl i ed t o t he di sci pl i ne of
moni t or i ng and di agnost i c syst ems [ M ann et al (1995)] . These t echni ques
ext end t h e power of t h e compu t er beyond t he usu al mat hemat i cal and
st at i st i cal f unct i ons by usi ng di al ogue and l ogi c t o det er mi n e var i ous
possi bl e cour ses of act i on or out co me. By pr ocessi ng i nf or mat i on much
f ast er (t han humans) t he t i me t o assess t he condi t i on and di agnose t he
causes of f ai l ur es can be r educed. It can anal yse si t uat i ons obj ect i vel y and
wi l l not f or get any r el evan t f act s (gi ven t hat i t has been suppl i ed t hem),
t her ef or e t he pr obabi l i t y of maki ng a wr ong assessment or di agnosi s may
be r educed. Fur t her mor e, i t can d et ect i nci pi ent f ai l ur es t hr o ugh i t s onl i ne
moni t or i ng of t he condi t i on par amet er s of t he syst em [ Laval l e et al (1993)] .
Impl ement at i on of condi t i on based mai nt enance
Havi ng i dent i f i ed and l i st ed al l t he condi t i on par amet er s of t he
mai nt enance si gni f i cant i t ems, t he ai m of t hi s st ep i s t o i mpl ement
condi t i on based mai nt enance. Accor di ng t o t he cl assi f i cat i ons of condi t i on
par amet er , condi t i on based mai nt enance coul d be di vi ded i n t w o pol i ci es:
Inspect i on Based Mai nt enance Pol i cy
5. M ai nt ai nabi l i t y and M ai nt enance 169
The sui t abl e mai nt enan ce pol i cy f or i t ems f or whi ch t hei r condi t i ons ar e
descr i b ed by t he r el evant condi t i on i ndi cat or , RCI i s i nspect i onbased
mai nt enance. The al gor i t hm, whi ch pr esent s t he mai n t enance pr ocedur e i n
t hi s case, i s show n i n Fi gur e 5.11
Inspect i on i s car r i ed at f i xed i nt er val s t o d et er mi ne whet h er t he
condi t i on of t he i t em, i s sat i sf act or y or unsat i sf act or y accor di ng t o t he RCI .
Bef or e t he i t em/ syst em i s i nt r odu ced i nt o ser vi ce t he most sui t abl e
f r equency of t h e i nsp ect i on, FMT
I
, and cr i t i cal val ue of r el evant condi t i on
i ndi cat or
cr
RCI has t o b e det er mi ned. Once t he cr i t i cal l evel i s r eached,
RCI FMT RCI
I
cr
( ) > , t he pr escr i b ed pr event i ve mai nt enance t asks t ake
pl ace. I f t he i t em f ai l s bet ween i nspect i ons, cor r ect i ve mai nt enance t akes
pl ace.
Advant ages of i nspect i on based mai nt enance
CBM has t he po t ent i al t o pr oduce l ar ge savi ngs si mpl y by al l ow i ng i t ems i n
t he syst em t o be r un t o t he end of t hei r usef ul l i f e. Thi s r educes t he
equi pment down t i me and mi ni mi ses bot h schedul ed and unsch edul ed
br eakdown si t uat i ons. By el i mi nat i ng al l unsched ul ed i nt er r u pt i ons t o
oper at i on and pr oduct i on and onl y car r yi ng out r equi r ed mai nt enance i n a
car ef ul l y cont r ol l ed mann er , i t i s possi bl e t o r educe t he mai nt enan ce cost ,
t o i mpr ove saf et y, i mpr o ve t he ef f i ci ency of t he oper at i on and i ncr ease t he
syst em’ s avai l abi l i t y.
5. M ai nt ai nabi l i t y and M ai nt enance 170
Fi gur e 5. 11 Al gor i t hm f or i nspect i on based mai nt enance t ask
The b en ef i t s of i nspect i on based mai nt enance pol i cy can be summar i sed as
f ol l ow s:
1. Reduce unplanned downtime, since maintenance engineers can determine
optimal maintenance intervals through the condition of constituent items
in the system. This allows for better maintenance planning and more
efficient use of resources.
2. Improve safety, since monitoring and detection of the deterioration in
condition and/or performance of an item/system will enable the user to
stop the system (just) before a failure occurs.
3. Extending the operating life of each individual items and therefore the
coefficient of life utilisation will be increased compared to time based
maintenance
4. Improve availability by being able to keep the system running longer and
reducing the repair time.
5. Reduce maintenance resources due to reduction in unnecessary
maintenance activities
6. The above benefits will lead to a reduction in maintenance costs
MAINTENANCE PROCEDURE
Conditionbased Maintenance policy
System in use
Maintenance task
Type Inspection
Inspection of RCI
Yes
No
Determination of
RCI > RCI
cr
cr
FMT and RCI
I
5. M ai nt ai nabi l i t y and M ai nt enance 171
Exami nat i on Based Mai nt enance Pol i cy
The d eci si on f or per f or mi ng t he condi t i onbased mai nt enan ce t asks i s
based on t he i nf or mat i on r el at ed t o t he condi t i on of an i t em/ syst em
est abl i shed t hr ough condi t i on checks dur i ng i t s o per at i onal l i f e. Thi s
i ndi cat es t hat i nsp ect i onbased mai nt enance st r at egy has achi eved t he
demand f or i ncr easi ng t he l evel of ut i l i sat i on of an i t em/ syst em. How ever ,
t he syst em avai l abi l i t y may not i ncr ease, due t o an i ncr eased number of
i nt er r upt i ons of t h e oper at i on caused by i ncr easi ng t he number of
i nspect i ons. Th er ef o r e, as an al t er nat i ve, exami nat i on based mai nt enance
appr oach i s pr oposed by Knezevi c (1987b) f or t he d et er mi nat i on of
mai nt enan ce t asks based on r el evant condi t i on pr edi ct or s.
Exami nat i on based mai nt enance pr o vi des addi t i onal i nf or mat i on about t he
change i n condi t i on of t he i t ems consi der ed dur i n g i t s oper at i onal l i f e.
Conseq uent l y, exami nat i on based mai nt enance was devel oped f or t he
cont r ol of mai nt enance pr ocedur es [ El  Har am 1995] . Wi t h mor e
i nf or mat i on about t he pr ocess of change i n condi t i on, a hi gher l evel of
ut i l i sat i on of t he i t ems can b e achi eved w hi l st mai nt ai ni ng a l ow pr obabi l i t y
of f ail ur e dur i ng t he oper at i on.
It i s a dynami c pr ocess becau se t he t i me of t he n ext exami nat i on i s f ul l y
det er mined b y t he r eal condit i on of t he syst em at t he t i me of exami nat i on.
Dynami c con t r ol of mai nt enance t asks al l ows each i ndi vi dual i t em t o
per f or m t he r equest ed f unct i on wi t h t he r equi r ed pr obabi l i t y of f ai l ur e, as
i n t he case of t i mebased pr even t i ve mai nt enance but wi t h f ul l er ut i l i sat i on
of oper at i ng l i f e, hen ce wi t h a r educt i on of t ot al cost of oper at i on and
pr oduct i on.
The cr i t i cal l evel of t h e r el evant condi t i on pr edi ct or RCP
cr
, set s t he l i mi t
above whi ch appr opr i at e mai nt enance t asks shoul d be per f or med. The
i nt er val b et w een t he l i mi t (
lim
RCP ) and cr i t i cal val u es d epends on t he
abi l i t y of t he oper at or t o measur e t h e condi t i on of t he i t em t hr ough t he
RCP. The i t em under consi d er at i on coul d be i n one of t he f ol l ow i n g t hr ee
st at es, accor di ng t o t he numer i cal val ue of t he RCP, :
1. RCP RCP l RCP
initial cr
< < ( ) : cont i nue wi t h exami nat i ons;
2. RCP RCP l RCP
cr
< < ( )
lim
: pr event i ve mai nt enance t ask r equi r ed;
3. RCP RCP l
lim
( ) < : cor r ect i ve mai nt enance t ask, because t h e f ai l ur e
has al r eady occur r ed.
5. M ai nt ai nabi l i t y and M ai nt enance 172
In or der t o mi ni mi se i nt er r upt i ons t o t he oper at i on and maxi mi se t he
avai l abi l i t y of t he syst em, no st oppages occur unt i l t he t i me t o t h e f i r st
exami nat i on of t he condi t i on of t he i t em, FMT
E
1
. The r esul t of t he
exami nat i on i s gi ven as a numer i cal val ue of t he r el evant condi t i on
pr edi ct or , MRCP FMT
E
( )
1
, and i t pr esent s t he r eal condi t i on of t he i t em
at t hi s i nst ant of t i me. The f ol l owi ng t wo condi t i ons ar e possi bl e,
dep endent on t he val ue r ecor ded:
1. MRCP FMT RCP
E
cr
( )
1
> , whi ch means t hat a pr escr i b ed
mai nt enance t ask shoul d t ake pl ace.
2. MRCP FMT RCP
E
cr
( )
1
< , t he i t em can cont i nue t o be used.
The quest i on, w hi ch i mmedi at el y ar i ses h er e, i s: w hen w i l l t he next
exami nat i on have t o be don e, pr eser vi ng t he r equi r ed r el i abi l i t y l evel ? The
t i me t o t he next exami nat i on dep ends on t he di f f er ence bet ween t he
RCP
cr
and MRCP FMT
E
( )
1
. The gr eat er t he di f f er ence, t he l onger t he
(oper at i onal ) t i me t o t he next exami nat i on, FMT
E
2
. At t he pr ed et er mi n ed
t i me of t he next exami nat i on, FMT
n
E
, ei t her of t he t wo condi t i ons i s
possi bl e, and t he same pr ocedur e sho ul d be f ol l ow ed, as show n i n Fi gur e
5. 12
Advant ages of Exami nat i on Based Pol i cy
The advant ages of t he exami nat i onbased mai nt enance pol i cy ar e:
1. Fuller utilisation of the functional life of each individual system than in
case of time based maintenance;
2. Provision of the required reliability level of each individual system as in
case of timebased maintenance;
3. Reduction of the total maintenance cost as a result of extending the
realisable operating life of the system and provision of a plan for
maintenance tasks from the point of view of logistic support;
4. Increased availability of the item by a reduction of the number of
inspections in comparison with inspectionbased maintenance.
5. M ai nt ai nabi l i t y and M ai nt enance 173
5. Applicability to all engineering systems. The main difficulties are the
selection of a relevant condition predictor and the determination of the
mathematical description of the RCP l ( ) .
Fi gur e 5.12 M ai nt enance pr ocedur e f or exami nat i on based mai nt enance
In pr act i ce, i t i s i mpossi bl e t o el i mi nat e al l br eakd ow ns. In some cases, i t
may no t b e economi cal or pr act i cal t o use exami nat i onbased mai nt enance.
Somet i mes i t i s not physi cal l y possi bl e t o moni t or t he condi t i on of al l
mai nt enan ce si gni f i cant i t ems. For t hese r easons, condi t i on based
mai nt enan ce shoul d not b e consi der ed t o be a st andal one pol i cy. It shoul d
be i nt egr at ed as a par t of t he over al l mai nt enan ce pol i cy. Thus, t he opt i mal
sel ect i on of mai nt enance pol i cy f or a syst em shoul d i ncl ude f ai l ur ebased,
cr
E
n
RCP FMT MRCP > ) (
Maintenance Policy
Condition Based Maintenance
Policy
Type
Examination
Determination of
E
FMT
1
and
cr
RCP
System in use
Examination of RCPat
E
FMT
1
cr
E
RCP FMT MRCP > ) (
1
Determination of
E
n
FMT
Preventive task
System in use
Examination of RCPat
E
n
FMT
5. M ai nt ai nabi l i t y and M ai nt enance 174
t i mebased, i nspect i on based and exami nat i onbased mai nt enance
st r at egi es. The r easo ns f or t hi s ar e summar i sed bel ow:
1. Not all items in the system are significant; the suitable maintenance
policy is therefore, failurebased maintenance.
2. It may not be possible or practical to monitor the condition / performance
of the significant items, so the suitable maintenance policy is therefore,
timebased maintenance.
3. If the condition parameters of a significant item cannot be described by a
relevant condition predictor, then the suitable maintenance policy is
inspectionbased maintenance
4. For significant items with relevant condition predictors, the most suitable
policy is examinationbased maintenance.
A mai nt enance management appr oach such as r el i abi l i t y cent r ed
mai nt enan ce coul d be used t o sel ect t he most appl i cabl e and ef f ect i ve
mai nt enan ce t ask f or each i t em i n t he syst em
37. MAINTENANCE RESOURCES
It i s i mpor t ant t o st r ess t hat t he number of act i vi t i es, t hei r sequen ce and
t he t yp e and quant i t y of r esour ce r equi r ed mai nl y depends on t he deci si ons
t aken dur i ng t he desi gn phase of t he i t em/ syst em. The t i me r equi r ed t o
per f or m a mai nt enance t ask wi l l al so depend on deci si ons made dur i ng t hi s
phase, such as t he compl exi t y, t est abi l i t y, accessi b i l i t y and any speci al
f aci l i t i es, equi pment , t ool s or r esour ces needed.
Resour ces r equi r ed pr i mar i l y t o f aci l i t at e t he mai nt enance pr ocess wi l l be
cal l ed M ai nt enance Resour ces, M R.. The r esour ces needed f or t he
successf ul co mpl et i on of ever y mai nt enance t ask, coul d be gr ouped i nt o
t he f ol l ow i ng cat egor i es (Kn ezevi c 1997):
1. Maintenance Supply Support, MSS: is generic name which includes all
spares, repair items, consumables, special supplies, and related
inventories needed to support the maintenance process
2. Maintenance Test and Support Equipment, MTE: includes all tools,
special condition monitoring equipment, diagnostic and checkout
equipment, metrology and calibration equipment, maintenance stands and
servicing and handling equipment required to support maintenance tasks
associated with the item/system. Typically, MTE can be divided into two
groups: special to type equipment (STTE) and general (to type)
equipment (GTTE).
3. Maintenance Personnel, MP: required for the installation, checkout,
handling, and sustaining maintenance of the item/system and its
5. M ai nt ai nabi l i t y and M ai nt enance 175
associated test and support equipment are included in this category.
Formal training for maintenance personnel required for each maintenance
task should be considered
4. Maintenance Facilities, MFC: refers to all special facilities needed for
completion of maintenance tasks. Physical plant, real estate, portable
buildings, inspection pits, dry dock, housing, maintenance shops,
calibration laboratories, and special repair and overhaul facilities must be
considered related to each maintenance task
5. Maintenance Technical Data, MTD: necessary for checkout procedures,
maintenance instructions, inspection and calibration procedures, overhaul
procedures, modification instructions, facilities information, drawings
and specifications that are necessary in the performance of system
maintenance functions. Such data not only cover the system but test and
support equipment, transportation and handling equipment, training
equipment and facilities
6. Maintenance Computer Resources, MCR: refers to all computer
equipment and accessories, software, program tapes/disks, data bases and
so on, necessary in the performance of maintenance functions. This
includes both condition monitoring and diagnostics.
On t he ot h er hand, i t i s i mpor t ant t o r em ember t hat each t ask i s per f or med
i n a speci f i c wor k envi r onment t hat coul d make a si gni f i cant i mpact on t he
saf et y, accur acy and ease of t ask compl et i on. Th e mai n envi r onment al
f act or s coul d be gr oup ed as f oll ows:
• space i mp edi m ent (whi ch r ef l ect s t he obst r uct i ons i mposed on
mai nt enance per sonn el dur i ng t he t ask execu t i on whi ch r equi r es t hem
t o oper at e i n aw kw ar d posi t i ons)
• Cl i mat i c condi t i ons such as r ai n/ snow, sol ar r adi at i on, humi di t y,
t emper at ur e, and si mi l ar si t uat i ons, whi ch coul d make si gni f i cant
i mpact on t he saf et y, accur acy and ease of t ask compl et i on.
• Pl at f or m on whi ch mai nt enance t ask i s per f or med (on oper at i onal si t e,
on boar d a shi p/ submar i ne, space vehi cl e, wor kshops, and si mi lar ).
38. MAINTENANCE INDUCED FAILURES
Wh enever t he cause of f ai l ur e i s r el at ed t o t he mai nt enance per f or med on
t he syst em, we cal l i t mai nt enancei nduced f ai l ur e M IF. The r oot cause of
M IF i s poor wor kmanshi p, whi ch mi ght l ead t o po or spar es or mat er i al
sel ect i on, i mpr oper use of t est equi pment , t r ai ni ng, wor ki ng envi r onment
5. M ai nt ai nabi l i t y and M ai nt enance 176
et c. A f ew exampl es of mai nt enancei nduced f ai l ur e ar e di scussed i n t hi s
sect i on.
In 1991, Ni gel M ansel l l ost hi s chance of beco mi ng t he For mul a 1 Wor l d
Champi o n i n Por t ugal when one of t he mechani cs dur i ng a r out i ne t yr e
change cr osst hr ead ed t he r et ai ni ng nut on t he r ear of f si de wheel . The
r esul t was t hat t he wh eel over t ook t he car as Ni gel was exi t i ng f r om t h e pi t
l ane and hi s chan ce of vi ct or y and of t h e champi o nshi p ended at t hat
moment .
An ai r l i ne pi l ot had a ver y l ucky escape when he was near l y sucked t hr ou gh
a wi ndow i n t he cockpi t . The wi ndow was r emoved and r epl aced dur i ng a
r ecent l y compl et ed mai nt enance act i vi t y. Wh en t h e cabi n was pr essur i sed
as t he ai r cr af t cl i mbed t o cr ui si ng al t i t ude, t he wi ndow bl ew out . The r api d
l oss of pr essur e caused t he pi l ot sat next t o t he wi ndow t o be sucked
t hr ough t h e hol e. A combi nat i on of hi s si ze and t he qui ck r eact i ons of ot her
member s of t h e cr ew w er e al l t hat saved hi m f r om a cer t ai n deat h. The
cause of t h e wi ndow bei ng bl own out was t hat i t had been r ef i t t ed usi ng
under si zed scr ew s.
In 1983, a n ew Ai r Canada Boei ng 767 f l yi ng f r om M ont r eal t o Edm ont on
r an out of f uel hal f w ay bet w een t he t w o at Gi ml i near Wi nni peg. Al t hough
t hi s was not ent i r el y t he f aul t of t h e r ef uel l er s, t hei r mi scal cul at i ons i n
conver t i ng bet w een i mper i al an d met r i c uni t s w as t he f i nal st r aw i n an
unf or t unat e sequence of event s. A number of r eco mmen dat i ons f ol l owed
t hi s i nci dent whi ch shoul d mean t hat i t n ever happens agai n (pr ovi ded
ever yone f ol l ow s t he pr ocedur es cor r ect l y).
A f ew year s ago, a t eam of “ exp er i enced ” mechani cs t hough t t hey knew
how t o do a par t i cul ar mai nt enance t ask so di d not f ol l ow t he i nst r uct i ons
i n t he mai nt enance manual s. The r esul t was a cost of sever al mi l l i on
pounds st er l i ng and a numb er of ai r cr af t bei ng out of ser vi ce f or
consi der abl y l onger t han t hey shoul d have been.
These ar e ext r eme exampl es of what may be consi der ed as “ mai nt enance
i nduced f ai l ur es” . They ar e al so ones w her e i t w as r el at i vel y easy t o
det er mi ne t he cause(s).
One of t he maj or causes f or acci dent al damage t o compon ent s (f r om l i ne
r epl aceabl e uni t s t o par t s) i s t he need t o r emove t hem i n or d er t o access
ot her component s. Usi ng CATIA an d EPI C (or si mi l ar syst ems) can do a
gr eat d eal t o ai d t he t ask of maki ng co mponent s accessi bl e and r emovi ng
i nt er f er ence pr ovi ded, of cour se, t he desi gn t eam ar e aw ar e of t h ese needs
and t hei r i mpor t ance t o t he oper at i onal ef f ect i veness of t he ai r cr af t .
Fast ener s not pr oper l y t i ght ened and l ocked (wher e appr opr i at e) can wor k
l oose. Si mi l ar l y, i f t hey ar e not “ capt ur ed” t h en t her e i s a danger of t hem
bei ng “ l ost ” w hen undone. If t hey ar e i nsi de t he engi ne or engi ne nacel l e
5. M ai nt ai nabi l i t y and M ai nt enance 177
t hey may b e sucked i nt o t h e d el i cat e machi ner y al most cer t ai nl y causi ng
ext ensi ve and expensi ve damage. Fast ener s over t i ght ened may cause
di st or t i on r esul t i ng i n l eaks or damage, whi ch may agai n have ser i ous
consequen ces. Consi st en t and sensi bl e use of f ast ener s can not onl y
r educe su ch pr obl ems but wi l l al so r educe t he par t s l i st and hence i mpr ove
t he suppor t abi l i t y of t he ai r cr af t .
Some spar e par t s may be expensi ve or di f f i cul t t o obt ai n. Ther e may be a
t empt at i on t o use al t er nat i ve sour ces (ot her t han t hose aut hor i sed). In
many cases t hese may be made f r o m i nf er i or mat er i al s or t o l ess
demandi ng t ol er ances and qual i t y st andar ds. The use of such r o gue par t s
may r esul t i n pr emat ur e component f ai l ur e and, possi bl e, ser i ous damage.
Conf igur at i on cont r ol and f ul l t r aceabi l i t y of par t s i s an essent i al el emen t of
ai r cr af t saf et y but , unt i l pr act i cal el ect r oni c t aggi ng of al l par t s becomes
avai l abl e, i t wi ll r emai n di f f i cul t t o pol i ce ef f ect i vel y.
39. MAINTENANCE COST
The wor l d' s ai r l i nes spend ar ound $21 bi l l i on on mai nt enance, out of whi ch
21% i s spen t on l i ne mai nt enance, 27% on heavy mai n t enance, 31% on
engi n e over haul , 16% on componen t o ver haul and t he r emai nder on
modi f i cat i ons and conver si ons (M Lam 1995). Repai r and mai nt enance of
bui l di ng st ock i n t he UK r epr esent s o ver 5% of Gr oss Domest i c Pr o duct , or
£36 bi l l i on at 1996 [ Bui l di ng mai nt enance i nf or mat i on r epor t 254,1996] .
M ai nt enance and r epai r cost s can be t wo t o t hr ee t i mes t h e i ni t i al capi t al
cost s, over t he l i f e of many t ypes of bui l di ngs.
If one r ecogni ses t hat mai nt enance i s essent i al l y t he managemen t of f ai l ur e
t hen cl ear l y, t hi s expendi t ur e i s pr i mar i l y t he r esul t of poor qual i t y and
unr el i abil i t y. How ever , si nce i t i s i mpossi bl e t o pr oduce a syst em w hi ch w i l l
never f ai l i f oper at ed f or l ong enough w e must consi der w ays i n w hi ch t he
cost s of mai nt enance can b e kept t o a mi ni mum w hi l st ensur i ng syst em
avai l abi l i t y, saf et y and i nt egr i t y.
We have al r eady seen t hat t her e ar e many f act or s whi ch can af f ect t he
cost s of mai nt ai ni ng a syst em. Whi l st t he or i gi nal d esi gn w i l l be a maj or
i nf l uenci ng f act or on t hese cost s, t he oper at or s and mai nt ai ner s of t he
syst em can, non et hel ess, do much t o mi ni mi se t h e cost of owner shi p by
adopt i ng t h e most sui t abl e mai nt enance pol i ci es f or t he condi t i ons
pr evai l i ng.
5. M ai nt ai nabi l i t y and M ai nt enance 178
39.1 Cost of Maintenance Task
The cost of t he mai nt enance t ask i s t he cost associ at ed wi t h each cor r ect i ve
or pr event i ve t ask, whet her t i mebased or condi t i onbased. The expect ed
cor r ect i ve mai nt enance cost i s t he t ot al cost of mai nt enance r esour ces
needed t o r epai r or r epl ace f ai l ed i t ems. Si mi l ar l y, t he expect ed pr event i ve
mai nt enan ce cost i s t h e t ot al cost of mai nt enan ce r esour ces needed t o
i nspect and/ or exami ne an i t em bef or e f ai l ur e t akes pl ace and t o r epl ace
any i t ems r ej ect ed. Thus, t he t ot al mai n t enance cost t hr oughout t he l i f e of
a syst ems/ pr oduct i s t he sum of t he cor r ect i ve and pr event i ve mai nt enance
cost s and t he o ver head cost s, w hi ch consi st of al l co st s ot her t han di r ect
mat er i al , l abour and pl an t equi p ment . Th e cost of mai nt enan ce t ask can be
di vi ded i nt o t wo cat egor i es:
39.2 Direct cost of maintenance task
The di r ect cost associ at ed wi t h each mai nt enance t ask, CM T, i s r el at ed t o
t he cost of mai nt enance r esour ces, CM R, whi ch ar e ment i on ed i n Sect i on 9.
Thi s i s t he cost of t he mai nt enance r esour ces di r ect l y used dur i ng t he
execut i on of t he mai nt enance t ask, w hi ch i s def i ned as:
d f te p m s
C C C C C C CMT + + + + + · (5.11)
Where:
s
C = cost of spare parts,
m
C = cost of material,
p
C = cost of
personnel,
te
C = cost of tools and support equipment,
f
C = cost of
facilities and
d
C = cost of technical data.
39.3 Indirect cost of maintenance task
Indi r ect cost s i ncl udes as management and admi ni st r at i on st af f need ed f or
t he su ccessf ul compl et i on of t he t ask and t he cost of t he consequences of
not havi ng t he syst em avai l abl e whi ch i s r el at ed t o a compl et e or par t i al
l oss of pr oduct i on and/ or r evenue. It al so i ncl udes t he over head cost s, i .e.
sal ar i es of empl oyer s, h eat i ng, i nsur ance, t axes, f aci l i t i es, el ect r i ci t y,
t el ephon e, IT, t r ai ni ng and si mi l ar w hi ch ar e i ncur r ed w hi l e t he i t em i s i n
st at e of f ai l ur e (and, of cour se, not i ncl uded i n t h e di r ect cost s). These cost s
shoul d not be negl ect ed, b ecause t hey coul d be even hi gher t han t he ot her
cost el ement s.
5. M ai nt ai nabi l i t y and M ai nt enance 179
Cost of l ost pr oduct i on and/ or r evenue, CLR, i s di r ect l y pr opor t i onal t o t he
pr oduct of t he l engt h of t he t i me whi ch t he syst em spends i n t he st at e of
f ai l u r e (dow n t i me) and t he i ncome hour l y r at e, IHR, w hi ch i s t he money
t he syst em w oul d ear n w hi l st i n oper at i on. Thus, t he cost of l ost r evenue
coul d be det er mi ned accor di ng t o t he f ol l ow i ng expr essi on:
IHR DT IHR DST DMT CLR × · × + · ) ( (5.12)
Wh er e DM T i s dur at i on of mai nt enance t ask, DST i s dur at i on of suppor t
t ask and DT i s t ot al down t i me. Not e f or syst ems t hat ar e not nor mal l y i n
cont i nuous oper at i on, t he downt i me shoul d t ake acco unt of t he pr opor t i on
of t he t i me t he syst em woul d nor mal l y be expect ed t o be oper at i onal . In
par t i cul ar , pr event at i ve, pl anned or sch edul ed mai nt enan ce woul d
nor mal l y be don e w hen t he syst em w oul d be exp ect ed t o be i dl e and w oul d
onl y count as “ downt i me” f or any per i od t hat t h e syst em woul d be
exp ect ed t o be oper at i onal . Thus, f or exampl e, i f an ai r l i n er i s not
per mi t t ed t o f l y b et w een t h e hour s of 21: 00 and 07:00 t hen any
mai nt enance t asks und er t aken and compl et ed dur i ng t hose 10 hour s woul d
not af f ect t he r evenueear ni n g capaci t y of t he ai r cr af t .
39.4 Total cost of maintenance task
The t ot al cost of mai nt enance t ask i s t he sum cost of di r ect and i ndi r ect
cost s, t hus:
CMT CMR CLR · + (5.13)
M aki ng use of t he above equat i ons t h e expr essi on f or t he cost of t he
compl et i on of each mai nt enance t ask i s def i ned as:
IHR DST DMT C C C C C C CMT
d f te p m s
× + + + + + + + · ) ( (5.14)
It i s necessar y t o under l i ne t hat t he cost d ef i ned by t he above expr essi on
coul d di f f er consi der abl y, due t o:
1. Adopt i on of di f f er ent mai nt enan ce pol i ci es
2. The di r ect cost of each mai nt enance t ask
3. Consump t i on of mai nt enance r esour ces
4. Dur at i on of mai nt enance t ask,
E I p c
DMT DMT DMT DMT and , ,
5. M ai nt ai nabi l i t y and M ai nt enance 180
5. Fr equency of pr event i ve mai nt enance t ask,
L
FMT , t he f r equency of
i nspect i on,
I
FMT and f r equency of exami nat i on
E
FMT
6. Dur at i on of suppor t t ask,
E I p c
DST DST DST DST and , ,
7. The exp ect ed number of mai nt enance t asks NMT T
st
( ) per f or med
dur i ng t h e st at ed oper at i onal l engt h,
st
L . For exampl e, i n t he case of
FBM ,
MTTF
T
T NMT
st
st
· ) (
8. Di f f er ent pr obabi l i t y di st r i but i ons and di f f er ent val ues whi ch r andom
var i abl es
E I p c E I p c
DST DST DST DST DMT DMT DMT DMT and , , , , , ,
can t ake.
9. I ndi r ect cost s of mai nt enance t asks.
Thus, t he gen er al expr essi on f or t he cost of each mai nt enance t ask w i l l
have di f f er ent dat a i nput f or di f f er ent mai nt enance pol i ci es, as show n
bel ow :
c c c c
d
c
f
c
te
c
p
c
m
c
sp
c
IHR DST DMT C C C C C C CMT × + + + + + + + · ) (
p p p p
d
p
f
p
te
p
p
p
m
p
sp
p
IHR DST DMT C C C C C C CMT × + + + + + + + · ) (
I I I I
d
I
f
I
te
I
p
I
m
I
sp
I
IHR DST DMT C C C C C C CMT × + + + + + + + · ) (
E E E E
d
E
f
E
te
E
p
E
m
E
sp
E
IHR DST DMT C C C C C C CMT × + + + + + + + · ) (
Wher e: CMT
c
i s r el at ed t o t he cost of each mai nt enance t ask per f or med
af t er t he f ai l ur e, CMT
p
i s cost i n t he case of t i me based mai nt enance
I
CMT i s cost of i nspect i on based mai nt enance and
I
CMT i s cost of
exami nat i on based mai nt enance.
The expect ed t ot al mai nt enance cost f or a st at ed t i me, ) (
st
T CMT , i s equal
t o t he pr oduct of t he mai nt enance cost f or each mai nt enance t ask and t he
expect ed number of mai nt enan ce t asks per f or med d ur i ng t he st at ed t i me,
NMT T
st
( ) , t hus:
) ( ) (
) ( ) ( ) (
st
E E
st
I I
st
p p
st
c c
st
T NMT CMT T NMT CMT
T NMT CMT T NMT CMT T CMT
× + ×
+ × + × ·
(5.15)
5. M ai nt ai nabi l i t y and M ai nt enance 181
39.5 Factor Affecting Maintenance Costs
M ai nt enance cost coul d be af f ect ed by t he f ol l owi ng f act or s:
1. Supply responsiveness or the probability of having a spare part available
when needed, supply lead times for given items, levels of inventory, and
so on.
2. Test and support equipment effectiveness, which is the reliability and
availability of test equipment, test equipment utilisation, system test
thoroughness, and so on.
3. Maintenance facility availability and utilisation.
4. Transportation times between maintenance facilities.
5. Maintenance organisational effectiveness and personnel efficiency.
6. Durability and reliability of items in the system
7. Life expectancy of system
8. Expected number of maintenance tasks
9. Duration of maintenance and support task
10. Maintenance task resources
In or der t o r educe mai n t enance cost s, i t i s necessar y t hat t he i mpact of t he
above f act or s shoul d be r educed and/ or cont r ol l ed.
In cal cul at i ng t he var i ous cost el ement s of mai nt enance, i t i s i mpor t ant t o
r eco gni se t hat f aci l i t i es, equi pment , and per sonn el may b e used f or ot her
t asks. For exampl e, mechani cs i n t he ar med f or ces may be put on guar d
dut y or pr ovi de a def ence r ol e when no t per f or mi ng mai nt enance t asks.
Thus el i mi nat i ng al l mai nt enance t asks at f i r st l i ne ( or OLevel ) may not
necessar i l y l ead t o a si gni f i cant r educt i on i n t he per sonnel depl oyed or ,
i ndeed, i n t he oper at i onal cost s.
40. AIRCRAFT MAINTENANCE  CASE STUDY
For ever y commer ci al ai r l i ne, mai nt enance i s on e of t he most i mpor t ant
f unct i ons t o assur e saf e oper at i on. Feder al Avi at i on Regul at i on (FAR)
r equi r e t hat , no per son may oper at e an ai r cr af t unl ess t h e mandat or y
r epl acement t i mes, i nspect i on i nt er val s and r el at ed pr ocedur es or
al t er nat i ve i nsp ect i on i nt er val s and r el at ed pr oced ur es set f or t h i n t he
oper at i ons sp eci f i cat i ons or i nspect i on pr ogr am has b een compl i ed w i t h. Al l
ai r cr af t must f ol l ow a mai nt enance pr ogr am t hat i s appr oved by a
r egul at or y aut hor i t y such as FAA (Feder al Avi at i on Admi ni st r at i on, USA) and
CAA (Ci vi l Avi at i on Aut hor i t y, UK). Each ai r l i ne d evel ops i t s own
mai nt enance pl an, based on t he manuf act ur er ’ s r ecommendat i ons and by
5. M ai nt ai nabi l i t y and M ai nt enance 182
consi d er i ng i t s ow n oper at i on. Thus, t w o di f f er ent ai r l i nes may have
sl i ght l y di f f er ent mai nt enance pr ogr am f or same ai r cr af t model used under
si mi l ar oper at i ng condi t i ons. Ai r cr af t mai nt enance i s r el i abi li t y cent r ed. It i s
cl ai med t hat each ai r cr af t r ecei ves appr oxi mat el y 14 hour s of mai nt enance
f or ever y hour i t f l i es (R Baker , 1995). M ai nt enance account s f or
appr oxi mat el y 10% of an ai r l i ne’ s t ot al cost s. On aver age a t ypi cal Boei ng
747 wi l l gener at e a t ot al ai r cr af t mai nt enan ce cost of appr oxi mat el y $1,700
per bl ock hour .
Ai r cr af t mai nt enance can be cat egor i sed as:
1. Rout i n e schedul ed mai nt enance.
2. Nonr out i ne mai nt enance.
3. Ref ur bi shment .
4. M odi f i cat i ons.
Rout i ne Schedul ed Mai nt enance
Schedul ed mai nt enance t asks ar e r equi r ed at d et er mi nant r ecur r i ng
i nt er val s or due t o Ai r wor t hi ness Di r ect i ves (AD). The most common
r out i ne mai nt enan ce i s vi sual i nspect i on of t he ai r cr af t pr i or t o a sch edul ed
depar t ur e (know n as w al k ar ound) by pi l ot s and mechani cs t o ensur e t hat
t her e ar e no obvi ous pr obl ems. Rout i ne mai nt enance can be cl assi f i ed as:
1. Over ni ght mai nt enan ce.
2. Har d t i me mai nt enance.
3. Pr ogr essi ve Inspect i on.
Over ni ght mai n t enance nor mal l y i ncl udes l ow l evel mai nt enance checks,
mi nor ser vi ci n g and speci al i nspect i ons done at t he end of t he wor ki ng f or
about one t o t wo hour s t o ensur e t hat t he pl ane i s oper at i ng i n accor dance
wi t h M i ni mum Equi pment Li st . Over ni ght mai nt enance pr ovi des an
oppor t uni t y t o r emed y passenger and cr ew compl ai nt s (M Lam, 1995).
Har d t i me i s t he ol d est pr i mar y mai nt enance pr ocess. Har d t i me r equi r es
per i odi c o ver haul or r epl acement of af f ect ed syst ems/ component s and
st r uct ur es and i s f l i ght , cycl e an d cal endar l i mi t ed. That i s, as soon as t he
5. M ai nt ai nabi l i t y and M ai nt enance 183
compon en t age r eaches i t har d t i me i t i s r epl aced w i t h a new co mponent .
M ost of t he r ot at i ng engi ne uni t s ar e har d t i med. The pur pose of har d t i me
mai nt enance i s t o assur e oper at i ng saf et y of component or syst em, whi ch
have a li mi t ed r edundancy.
Pr ogr essi ve i nspect i on gr o ups l i ke t i me r el at ed mai nt enance t asks i nt o
conveni ent ‘ bl ocks’ so t hat mai nt enance wor kl oad becomes bal anced wi t h
t i me and mai nt enance can be accompl i shed i n smal l ‘ bi t es’ maki ng
equi pment mor e avai l abl e. Gr oupi ng mai nt enance t asks al so hel ps bet t er
ut i l i sat i on of t he mai nt enance f aci l i t i es. These mai nt enance t ask gr oups ar e
(det ai l ed i nf or mat i on can be f ound i n M Lam (1995) and L R Cr awf or d,
1995):
1. Preflight Visual inspections carried out by the mechanic and the pilots
to ensure that there are no obvious problems.
2. A Check Carried out approximately every 150 flight hours, which
includes selected operational checks (general inspection of the
interior/exterior of the aircraft), fluid servicing, extended visual
inspection of fuselage exterior, power supply and certain operational
tasks. During A check, the aircraft is on ground for approximately 8 to
10 hours and requires approximately 60 labour hours.
3. B Check Occurs about every 750 flight hours and includes some
preventive maintenance such as engine oil spectroanalysis, oilfilter are
removed and checked, lubrication of parts as required and examination of
airframe. Also incorporates Acheck. The aircraft could be on ground for
10 hours and will require approximately 200 labour hours.
4. C Check Occurs every 3, 000 flight hour (approximately 15 months)
and includes detailed inspection of airframe, engines, and accessories. In
addition, components are repaired, flight controls are calibrated, and
major internal mechanisms are tested. Functional and operational checks
are also performed during Ccheck. It also includes both A and B
checks. The aircraft will be on ground for 72 hours and will require
approximately 3,000 labour hours.
5. D Check This is the most intensive form of routine maintenance occurs
about 20,000 flight hours (six to eight years). It is an overhaul that
returns the aircraft to its original condition, as far as possible. Cabin
interiors including seats, galleys, furnishings etc are removed to allow
careful structural inspections. The aircraft is on ground for about 30 days
and will require approximately 20,000 labour hours.
A and B checks and over ni ght mai nt enan ce ar e i nst ances of l i ne
mai nt enan ce (per f or med upon t h e ai r cr af t i nci den t al t o i t s sch edul ed
r evenue oper at i ons), of t en car r i ed out an ai r por t . C and D ch ecks, however
5. M ai nt ai nabi l i t y and M ai nt enance 184
ar e heavy mai nt enance t hat r equi r es speci al f acil i t ies and ext ensi ve l abour .
The t ask i nt er val s f or var i ous ch ecks men t i oned above coul d var y
si gni f i cant l y. The r ecommended t i me i nt er val s f or di f f er ent ai r cr af t mod el s
ar e gi ven i n Tabl e 5.3 (Ai r cr af t Economi cs).
Tabl e 5.3 Di f f er ent sch edul ed checks i n a commer ci al ai r cr af t
Ai r cr af t Type A check
Fl i ght hour s
B Check
Fl i ght hour s
C Check
Fl i ght hour s
D Check
Fl i ght hour s
Bo ei ng 707 90 450 14,000
Boei ng 727 80 400 1,600 16,000
Boei ng 737100 125 750 3,000 20,000
Bo ei ng 747100 300 3,600 25,000
DC8 150 540 3,325 23,745
DC9 130 680 3,380 12,600
Nonr out i ne mai nt enance r ef er s t o t he mai nt enance t asks t hat has t o be
per f or med on r egul ar basi s dur i ng checks, but whi ch i s not speci f i ed as
r out i ne mai nt enan ce t ask on t he j o b car ds of t he mai nt enance sch edul e.
Nonr out i ne mai nt enance shoul dn’ t be conf used wi t h unschedul ed
mai nt enan ce, whi ch i s r epai r s t hat have t o be do ne as a r esul t of an
unexpect ed f ai l ur e such as acci dent al damage (such as bi r d st r i ke) t o cr i t i cal
compon en t s or a r esponse t o ai r wor t hi ness di r ect i ves (AD). As t he ai r cr af t
age, t hey r equi r e mor e mai nt enance due t o f at i gue and cor r osi on. The
most si gni f i cant of t hese agi ng ai r cr af t ai r w or t hi ness di r ect i ves con cer ns
Bo ei ng 747. Th e f usel age of t he Bo ei ng 747 i s bui l t i n sect i ons as separ at e
ent i t i es and t hen assembl ed dur i ng t he ai r cr af t pr o duct i on phase. The
f usel age i s bui l t i n f i ve sect i ons and t he poi nt s at whi ch t hese sect i ons ar e
j oi ned ar e cal l ed t he pr oduct i on br eaks. Sect i on 41 i s t he sect i on f r om t he
nose t o j ust af t of t h e f or war d passenger ent r y (M ai nt ai ning t he Boei ng
747, Ai r cr af t Economi cs, 1994). The modi f i cat i on of Sect i on 41, w hi ch i s t he
ar ea ahead of t he f or war d passenger door s, r equi r es appr oxi mat el y 60,000
70,000 manhour s t o co mpl et e and r equi r es r epl acement of most of t he
st r uct ur al component s (L Cr aw f or d, 1995).
5. M ai nt ai nabi l i t y and M ai nt enance 185
5. M ai nt ai nabi l i t y and M ai nt enance 186
A v ai l ab i l i t y , D esi gn
f o r Rel i ab i l i t y an d
D at a A n al y si s
Co u r se M at er i al
Cour se Inst r uct or : Pr of essor U Di nesh Kumar
Indi an Inst i t ut e of M anagem ent Bangal or e
10. Avai l abi l i t y 187
Chapter 10
Availability
Ther e i s not hi ng i n t hi s wor l d const ant , but i nconsi st ency
Jonat han Sw i f t
Avai l abi l i t y i s used t o measur e t h e combi n ed ef f ect of r el i abi l i t y,
mai nt enance and l o gi st i c suppor t on t h e oper at i onal ef f ect i veness of t he
syst em. A syst em, w hi ch i s i n a st at e of f ai l ur e, i s not ben ef i ci al t o i t s
ow ner ; i n f act , i t i s pr obabl y cost i ng t he ow ner mo ney. If an ai r cr af t br eaks
down, i t cannot b e used un t i l i t has been decl ar ed ai r wor t hy. Thi s i s l i kel y
t o cause i nconveni ence t o t he cust omer s w ho may t h en d eci de t o sw i t ch t o
an al t er nat i ve ai r l i ne i n f ut ur e. It may di sr upt t he t i met abl es and cause
pr obl ems f or sever al days.
As ment i on ed i n Chap t er 9, most l ar ge ai r l i ner s have a ver y hi gh ut i l i sat i on
r at e w i t h t he onl y dow n t i me bei ng t o do a t r ansi t ch eck, unl oad, cl ean t he
cabi n, r ef uel , r est ock wi t h t he next f l i ght ’ s f oods and ot her i t ems, and
r el oad wi t h t he next set of passen ger s and baggage. The whol e op er at i on
gen er al l y t akes about an h our . Any del ay may cause i t t o mi ss i t s t ake of f
sl ot and mor e si gni f i cant l y i t s l andi ng sl ot , si nce an ai r cr af t cannot t akeof f
unt i l i t has been cl ear ed t o l and, even t hough t hi s may b e 12 h our s l at er .
M any ai r por t s cl ose dur i ng t he ni gh t t o avoi d unaccept abl e l evel s of noi se
pol l ut i on. If t he par t i cul ar f l i ght was due t o l and j ust bef or e t he ai r por t
cl oses, mi ssi ng i t s sl ot coul d mean a del ay of sever al hour s.
An oper at or of a syst em woul d l i ke t o make sur e t hat t he syst em wi l l be i n a
st at e of f unct i oni ng (SoFu) when i t i s r equi r ed. Desi gner s and
manuf act ur er s know t hat t hey ar e unl i kel y t o r emai n i n busi ness f or ver y
l ong i f t hei r syst ems do no t sat i sf y t he cust omer s’ r equi r ement s i n t er ms of
oper at i onal ef f ect i ven ess. M any f or ms of avai l abi l i t y ar e used t o measur e
t he ef f ect i veness of t he syst em. I nher ent avai l abi l i t y, oper at i onal
avai l abi l i t y and achi eved avai l abi l i t y ar e som e of t he measu r es used t o
10. Avai l abi l i t y 188
quant i f y wh et her an i t em i s i n an oper abl e st at e when r equi r ed. Avai l abi l i t y
i s def i ned as:
The probability that an item is in state of functioning at a given point
in time (point availability) or over a stated period of time (interval
availability) when operated, maintained and supported in a
prescribed manner.
It i s cl ear f r om t he above d ef i ni t i on t hat avai l ab i l i t y i s a f unct i on of
r el i abi l i t y, mai nt ai nabi l i t y and suppor t abi l i t y f act or s (Fi gur e 10.1).
Fi gur e 10.1 Avai l abi l i t y as a f unct i on of r el i abil i t y, mai nt ai nabi l i t y and
suppor t abi l i t y
In t hi s chapt er , w e l ook at f ew i mpor t ant avai l abi l i t y measur es such as poi nt
avai l abi l i t y, i nt er val avai l abi l i t y, st eady st at e i nher ent avai l abi l i t y,
oper at i onal avai l abi l i t y and achi eved avai l abi l i t y.
10.41. POINT AVAILABILITY
Poi nt avai l abi l i t y i s def i ned as t he pr obabi l i t y t hat t h e syst em i s i n t he st at e
of f unct i o ni ng (SoFu) at t he gi ven i nst ant of t i me t . We use t he no t at i on
A(t ) t o r epr esent t he poi nt avai l abi l i t y. Avai l abi l i t y expr essi ons f or syst ems
can be obt ai ned by usi ng st o chast i c pr ocesses. Dependi ng on t he t i me t o
f ai l ur e and t i me t o r epai r di st r i but i ons, one can use M ar kov chai n, r enewal
pr ocess, r egener at i ve pr ocess, semi M ar kov pr ocess and semi r egen er at i ve
pr ocess model s t o der i ve t h e expr essi on f or poi nt avai l abi li t y. For exampl e,
consi der an i t em wi t h const ant f ai l ur e r at e λ and const ant r epai r r at e µ. At
any i nst ant of t i me, t he i t em can be i n ei t her t he st at e of f unct i oni ng (say,
st at e 1) or i n t he st at e of f ai l ur e (say, st at e 2). As bot h f ai l ur e and r epai r
Rel i abi l i t y
M ai nt ai nabi li t y
Suppor t abi l i t y
Avai l abi l i t y
10. Avai l abi l i t y 189
r at es ar e const ant (and t hus f ol l ow exponent i al di st r ibut i on), we can use a
M ar kov chai n t o model t he syst em t o der i ve t he avai l abi l i t y expr essi on.
Let p
i j
( h) deno t e t he t r ansi t i on pr obabi l i t y f r om st at e i t o st at e j dur i ng t he
i nt er val ‘ h’ (i ,j = 1, 2). Def i ne, P
i
(t +h), as t he pr obab i l i t y t hat t he syst em
w oul d be i n st at e i at t i me t +h, f or i = 1, 2. The expr essi on f or P
1
(t +h) can be
der i ved usi ng t he f ol l owi ng l ogi c:
1. The syst em w as i n st at e 1 at t i me t and cont i nues t o r emai n i n
st at e 1 t hr ou ghout t he i nt er val h.
2. The syst em w as i n st at e 2 at t i me t and i t t r ansi t s t o st at e 1 d ur i ng
t he i nt er val h.
The cor r espondi ng expr essi on can be w r i t t en as:
) ( ) ( ) ( ) ( ) (
21 2 11 1 1
t p t P h p t P h t P × + × · + (10.1)
Usi ng si mi l ar l ogi c, t he expr essi on f or P
2
(t +h) can be wr i t t en as:
) ( ) ( ) ( ) ( ) (
22 2 12 1 2
h p t P h p t P h t P × + × · + (10.2)
p
1 1
(h) i s t he pr obabi l i t y of r emai ni ng i n st at e 1 dur i ng t h e i nt er val h. The
pr obabi l i t y p
11
(h) i s gi ven by
1 ) exp( ) (
11
h h h p λ λ − ≈ − · for λh<<1
p
2 1
(h) i s t he pr obabi l i t y of ent er i n g st at e 1 f r om st at e 2 dur i ng t he i nt er val
h. The cor r esp ondi ng expr essi on i s gi ven by
h h h p µ µ ≈ − − · ) exp( 1 ) (
21
for hµ<<1
p
1 2
(h) i s t he pr obabi l i t y of ent er i n g st at e 2 f r om st at e 1 dur i ng t he i nt er val
h. The pr obabi li t y p
12
(h) i s gi ven by
h h h p λ λ ≈ − − · ) exp( 1 ) (
12
for hλ<<1
p
2 2
(h) i s t he pr obabi l i t y of r emai ni ng i n st at e 2 dur i ng t h e i nt er val h. The
pr obabi l i t y p
22
(h) i s gi ven by:
h h h p µ µ − ≈ − · 1 ) exp( ) (
22
for hµ<<1
10. Avai l abi l i t y 190
Subst i t ut i ng t he val ues of p
i j
(h) i n equat i on (10.1) and (10.2), we get
h t P h t P h t P µ λ × + − × · + ) ( ) 1 ( ) ( ) (
2 1 1
) 1 ( ) ( ) ( ) ( ) (
2 1 2
h t P h t P h t P µ λ − × + × · +
By r ear r angi ng t he t er ms and set t i ng h → 0, w e have
) ( ) (
) ( ) ( ) (
2 1
1 1 1
0
t P t P
dt
t dP
h
t P h t P
Lt
h
µ λ + − · ·
− +
→
) ( ) (
) ( ) ( ) (
2 1
2 2 2
0
t P t P
dt
t dP
h
t P h t P
Lt
h
µ λ − · ·
− +
→
On sol vi ng t he above t wo di f f er ent i al equat i ons, we get
) ) ( exp( ) (
1
t t P µ λ
µ λ
λ
µ λ
µ
+ − ×
+
+
+
·
P
1
(t ) i s not hi ng but t h e avai l abi l i t y of t he i t em at t i me t , t hat i s t he
pr obabi l i t y t hat t he i t em wi l l be i n st at e of f unct i oni ng at t i me t . Thus, t he
poi nt avai l abi l i t y A(t ) i s gi ven by:
) ) ( exp( ) ( t t A µ λ
µ λ
λ
µ λ
µ
+ − ×
+
+
+
· (10.3)
Subst i t ut i ng λ = 1/ M TTF and µ = 1/ M TTR i n t he above equat i on, w e get
) )
1 1
( exp( ) ( t
MTTR MTTF MTTR MTTF
MTTR
MTTR MTTF
MTTF
t A + − ×
+
+
+
· (10.4)
Wh en t he t i me t o f ai l ur e and t i me t o r epai r ar e not exponent i al , we can use
a r egener at i ve pr ocess t o d er i ve t he avai l abi l i t y expr essi on. If f (t ) and g(t )
r epr esen t t he t i met of ai l ur e and t i met or epai r di st r i but i ons r espect i vel y,
t hen t h e poi nt avai l abi l i t y A(t ) can be wr i t t en as (Bi r ol ini , 1997):
dx x t F x g x f t F t A
t
n
n
)] ( 1 [ )] ( * ) ( [ ) ( 1 ) (
0 1
− − + − ·
∫ ∑
∞
·
10. Avai l abi l i t y 191
where [f(x)*g(x)]
n
is the nfold convolution of f(x)*g(x). The summation
∑
∞
·
∗
1
)] ( ) ( [
n
n
x g x f gives the renewal points f(x)*g(x), f(x)*g(x)*f(x)*g(x),
lies in [x, x+dx], and 1 F(tx) is the probability that no failures occur in the
remaining interval [x, t].
10.41.1 Average Availability
Int er val avai l abi l i t y, AA(t ), i s def i ned as t he expect ed f r act i onal dur at i on of
an i nt er val (0, t ] t hat t he syst em i s i n st at e of f unct i oni ng. Thus,
∫
·
t
dx x A
t
t AA
0
) (
1
) ( (10.5)
wher e A( x) i s t he poi nt avail abi l i t y of t he i t em as d ef i ned i n equat i on (10.3)
and (10.4). For an i t em w i t h const ant f ai l ur e r at e λ and const ant r epai r r at e
µ, t he aver age avai l abi l i t y i s gi ven by:
)] ) ( exp( 1 [
) (
) (
2
t
t
t AA µ λ
µ λ
λ
µ λ
µ
+ − −
+
+
+
· (10.6)
10.41.2 Inherent Availability
Inher ent avai l abi l i t y (or st ead yst at e avai l abi l i t y), A
i
, , i s def i ned as t he
st eady st at e pr obabi l i t y (t hat i s, t → ∞) t hat an i t em w i l l be i n a st at e of
f unct i oni ng, assumi ng t hat t hi s pr obabi l i t y depends onl y on t he t i met o
f ai l ur e and t i me t o r epai r di st r i but i ons. I t i s assumed t hat an y suppor t
r esour ces t hat ar e r equi r ed ar e avai l able w i t hout an y r est r i ct i on. Thus, t he
i nher ent avai l abi l i t y i s gi ven by:
MTTR MTTF
MTTF
t A Lt A
t
i
+
· ·
∞ →
) ( (10.7)
The above r esul t i s val i d f or any t i me t o f ai l ur e f unct i on F(t ) and any t i me t o
r epai r di st r i but i on G(t ) (Bi r ol i ni , 1997). Al so, i n t he case of const ant f ai l ur e
r at e λ and const ant r epai r r at e µ, t he f ol l owi ng i nequal i t y i s t r ue.
10. Avai l abi l i t y 192
) / exp( ) ( MTTR t A t A
i
− ≤ − (10.8)
Exampl e 10.1
Ti me t o f ai l ur e di st r i but i on of a di gi t al engi ne cont r ol uni t (DECU) f ol l ow s an
expon ent i al di st r i but i on wi t h mean t i me bet ween f ai l ur es 1200 hour s and
t he r epai r t i me al so f ol l ows an exponent i al di st r i but i on wi t h mean t i me t o
r epai r 400 hour s.
1. Pl ot t he poi nt avai l abi l i t y of t he DECU.
2. Fi nd t he aver age avai l abi l i t y of t he DECU dur i ng f i r st 5000 hour s.
3. Fi nd t he i nher ent avai l abi l i t y.
SOLUTION:
1. The poi nt avai l abi l i t y of t he DECU i s cal cul at ed usi ng t he equat i on
(10.4). Fi gur e 10.2 depi ct s t he poi nt avai l abi l i t y of t he syst em.
Fi gur e 10.2 Poi nt avai l abi l i t y of DECU
2. The aver age avai l abi l i t y of t he syst em dur i ng 5000 hour s of oper at i on i s
gi ven by:
)] ) ( exp( 1 [
) (
) (
2
t
t
t AA µ λ
µ λ
λ
µ λ
µ
+ − −
+
+
+
·
Subst i t ut i ng t he val ues of λ (= 1/ 1200) and µ (=1/ 400), we get t h e val ue
of t he aver age avai l abi l i t y dur i ng 5000 hour s as 0.7649.
0
0.2
0.4
0.6
0.8
1
0 1000 2000 3000 4000 5000 6000
Time
A
v
a
i
l
a
b
i
l
i
t
y
St eadyst at e
10. Avai l abi l i t y 193
3. The i nher ent avai l abi l i t y i s gi ven by
75 . 0
400 1200
1200
·
+
·
+
·
MTTR MTTF
MTTF
A
i
Thus, t he st ead y st at e avai l abi l i t y of t he syst em i s 0.75 or 75%.
10.41.3 System Availability of different reliability block
diagrams
Avai l abi l i t y of a syst em wi t h ser i es r el i abi l i t y bl ock di agr am wi t h n i t ems i s
gi ven by
) ( ) (
1
t A t A
n
k
i s
∏
·
· (10.9)
wher e A
i
(t ) i s t he poi nt avai l abi l i t y of i t h i t em. The i nher ent avai l abi l i t y of
t he syst em i s gi ven by
∏
·
+
·
n
k i i
i
s i
MTTR MTTF
MTTF
A
1
,
(10.10)
For a ser i es syst em w i t h al l t he el ement s havi ng const ant f ai l ur e and r epai r
r at es, t he syst em i nher ent avai l abi l i t y
s s
s
s i
MTTR MTTF
MTTF
A
+
·
,
(10.11)
M TTF
s
and M TTR
s
ar e syst em mean t i me t o f ai l ur e and syst em mean t i me t o
r epai r r espect i vel y. Let λ
i
and µ
i
r epr esent t he f ai l ur e r at e and r epai r r at e
of i t em i r espect i vel y. M TTF
s
and M TTR
s
ar e gi ven by
∑
·
·
n
i
i
s
MTTF
1
1
λ
∑
·
·
n
i
s
i i
s
MTTR
MTTR
1
λ
λ
, where ∑
·
·
n
i
i s
1
λ λ
10. Avai l abi l i t y 194
Avai l abil i t y of a par al l el syst em w i t h n i t ems i s gi ven by
∏
·
− − ·
n
i
i s
t A t A
1
)] ( 1 [ 1 ) ( (10.12)
Exampl e 10.2
A ser i es syst em consi st s of f our i t ems. Th e t i me t o f ai l ur e and t he t i me t o
r epai r di st r i but i ons of t he di f f er ent i t ems ar e gi ven as gi ven i n Tabl es 10.1
and 10. 2. Fi nd t he i nher ent avai l abi l i t y of t he syst em.
Table 10.1. Time to failure distribution for different items.
Item Number Distribution Parameters
Item 1 Weibull η = 2200 hours β = 3.7
Item 2 Exponential λ = 0.0008 per hour
Item 3 Weibull η = 1800 hours β = 2.7
Item 4 Normal µ = 800 hours σ = 180 hours
Table 10.2. Time to repair distribution for different items
Item number Distribution Parameters
Item 1 Lognormal µ
l
= 3.25 and σ
l
= 1.25
Item 2 Normal µ = 48 hours σ = 12 hours
Item 3 Lognormal µ
l
= 3.5 and σ
l
= 0.75
Item 4 Normal µ = 72 hours σ = 24 hours
SOLUTION:
Fi r st w e cal cul at e M TTF
i
and M TTR
i
f or di f f er ent i t ems:
10. Avai l abi l i t y 195
4 . 1984 902 . 0 2200 )
7 . 3
1
1 ( 2200 )
1
1 (
1
· × · + Γ × · + Γ × ·
β
η MTTF
1250 0008 . 0 / 1 / 1
2
· · · λ MTTF , 2 . 1600
3
· MTTF , MTTF
4
=800
33 . 56 ) 2 / exp(
2
1
· + ·
l l
MTTR σ µ hours, MTTR
2
=48 hours
87 . 43 ) 2 / exp(
2
3
· + ·
l l
MTTR σ µ hours, MTTR
4
= 72 hours
Inher ent avai l abi l i t y, A
i
, f or i t em i can be cal cul at ed usi ng t he equat i on
(10.11). Subst i t ut i ng t he val ues of M TTF
i
and M TTR
i
i n equat i on (10.11), w e
have
A
1
= 0.9723, A
2
= 0.9630, A
3
= 0.9733, A
4
= 0. 9174
The syst em avai l abi l i t y i s gi ven by
8362 . 0
4
1
· ·
∏
· i
i s
A A
10.42. ACHIEVED AVAILABILITY
Achi eved avai l ab i l i t y i s t he pr obabi l i t y t hat an i t em wi l l be i n a st at e of
f unct i oni ng (SoFu) wh en used as speci f i ed t aki ng i nt o account t he
schedul ed and unschedul ed mai nt enance; any suppor t r esour ces need ed
ar e avai l abl e i nst ant aneousl y. Achi eved avai l abi l i t y, A
a
, i s gi ven by
AMT MTBM
MTBM
A
a
+
· (10.13)
M TBM i s t he mean t i me b et ween mai nt enance and AM T i s act i ve
mai nt enance t i me. The mean t i me bet ween mai nt en ance dur i ng t he t ot al
oper at i onal l i f e, T, i s gi ven by:
sm
T T T M
T
MTBM
/ ) ( +
· (10.14)
M (T) i s t he r enew al f unct i on, t hat i s t h e expect ed number of f ai l ur es dur i ng
t he t ot al l i f e T. T
sm
i s t he sch edul ed mai nt enance i nt er val (t i me bet w een
schedul ed mai nt enance). The abo ve expr essi on i s val i d when af t er each
schedul ed mai nt enance, t he i t em i s ‘ asbadasol d’ and af t er each
10. Avai l abi l i t y 196
cor r ect i ve mai nt enance t he i t em i s ‘ asgoodasnew’ . The act i ve
mai nt enan ce t i me, AM T, i s gi ven by:
sm
sm
T T T M
MSMT T T MTTR T M
AMT
/ ) (
) / ( ) (
+
+ ×
· (10.15)
M TTR st ands f or t he mean t i me t o r epai r and M SM T i s t he mean sch edul ed
mai nt enan ce t ime.
Exampl e 10.3
Time to failure distribution of an engine monitoring system follows a
normal distribution with mean 4200 hours and standard deviation 420
hours. The engine monitoring system is expected to last 20,000 hours
(subject to corrective and preventive maintenance). A scheduled
maintenance is carried out after every 2000 hours and takes about 72
hours to complete the task. The time to repair the item follows a
lognormal distribution with mean time to repair 120 hours. Find the
achieved availability for this system.
SOLUTION:
Mean time between maintenance, MTBM, is given by
2000 / 20000 ) 20000 (
20000
/ ) ( +
·
+
·
M T T T M
T
MTBM
sm
M (20000) f or nor mal di st r i but i on wi t h mean 4200 hour s and st andar d
devi at i on 420 hour s i s gi ven by
1434 . 4 )
420
4200 20000
( ) 20000 (
1
·
×
× −
Φ ·
∑
∞
· n
n
M
n
hours 1414
10 1434 . 4
20000
≈
+
· MTBM
The act i ve mai nt enance t i me i s gi ven by:
10. Avai l abi l i t y 197
06 . 86
10 1434 . 4
72 10 120 1434 . 4
/ ) (
) / ( ) (
≈
+
× + ×
·
+
+ ×
·
sm
sm
T T T M
MSMT T T MTTR T M
AMT
The achi eved avai l abi l i t y of t he syst em i s gi ven by:
9426 . 0
06 . 86 1414
1414
·
+
·
+
·
AMT MTBM
MTBM
A
a
10.43. OPERATIONAL AVAILABILITY
Oper at i onal avai l abi l i t y i s t he pr obabi l i t y t hat t he syst em w i l l be i n t he st at e
of f unct i oni ng (SoFu) when used as sp eci f i ed t aki ng i nt o account
mai nt enan ce and l ogi st i c del ay t i mes. Oper at i onal avai l abi l i t y, A
o
, i s gi ven
by
DT MTBM
MTBM
A
o
+
· (10.16)
wher e, M TBM i s t he mean t i me bet ween mai nt enance (i ncl udi ng bot h
schedul ed and unschedul ed mai nt enance) and DT i s t he Down t i me. The
mean t i me b et w een mai nt enan ce dur i ng t he t ot al oper at i onal l i f e, T, i s
gi ven by:
sm
T T T M
T
MTBM
/ ) ( +
· (10.17)
M (T) i s t he r enew al f unct i on, t hat i s t h e expect ed number of f ai l ur es dur i ng
t he t ot al l i f e T. T
sm
i s t he schedul ed mai nt enance i nt er val (t i m e bet w een
schedul ed mai nt enan ce). The syst em down t i me DT i s gi ven by:
sm
sm
T T T M
MSMT T T MTTRS T M
DT
/ ) (
) / ( ) (
+
+ ×
· (10.18)
M TTRS st ands f or t he mean t i me t o r est or e t h e syst em and M SM T i s t he
mean schedul ed mai nt enance t i me. MTTRS i s gi ven by
10. Avai l abi l i t y 198
M TTRS = M TTR + M LDT
wher e MLDT i s t he mean l ogi st i c del ay t i me f or suppl y r esour ces. In t he
absence of any sch edul ed mai nt enance t he oper at i o nal avai l abi l i t y can be
cal cul at ed usi ng t he f ol l ow i ng si mpl e f or mul a
MLDT MTTR MTBF
MTBF
A
O
+ +
· (10.19)
Exampl e 10.4
In t he pr evi ous exampl e, assume t hat whenever a syst em f ai l s i t t akes
about 48 hour s bef or e al l t he necessar y suppor t r esour ces ar e avai l ab l e.
Fi nd t he oper at i onal avai l abi l i t y.
SOLUTION
M TBM i s same as i n t he pr evi ous exampl e and i s equ al t o 1414 hour s. The
mean t i me t o r est or e t he syst em i s gi ven by
M TTRS = M TTR + M LDT = 120 + 48 = 168 hour s
The syst em down t i me i s gi ven by
hours 12 . 100
1434 . 14
72 10 168 1434 . 4
/ ) (
) / ( ) (
·
× + ×
·
+
+ ×
·
sm
sm
T T T M
MSMT T T MTTRS T M
DT
The oper at i onal avai l abi l i t y of t he syst em i s gi ven by
9338 . 0
12 . 100 1414
1414
·
+
·
+
·
DT MTBM
MTBM
A
O
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 199
Chapter 11
Design for Reliability, Maintenance and
Logistic Support
Rel i abi l i t y, mai nt enance and suppor t abi l i t y shoul d be desi gned i nt o t he
pr oduct . Desi gn phase i s par t i cul ar l y i mpor t ant f or any pr oduct as t he
deci si ons mad e dur i n g t hi s st age can det er mi ne how r el i abl e t he pr oduct i s
goi ng t o be as wel l as t he mai nt ai n abi l i t y and suppor t abi l i t y of t hat pr oduct .
In t hi s chapt er , we woul d l i ke t o di scuss f ew t ool and t echni qu es t hat can
be used at t he d esi gn st age t o i mpr ove t he RM S char act er i st i cs.
44. RELIABILITY ALLOCATION
Rel i abi l i t y al l ocat i on i s a pr ocess by w hi ch t he syst em’ s r el i abi l i t y
r equi r ement s i s di vi ded i nt o subsyst em and component r el i abi l i t y
r equi r ement s.
45. FAILURE MODES, EFFECTS AND CRITICALITY
ANALYSIS (FMECA)
The f ai l ur e modes, ef f ect s and cr i t i cal i t y anal ysi s (FM ECA) i s a syst emat i c
met hod f or exami ni ng al l modes t hr ough whi ch a f ai l ur e can occur ,
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 200
pot ent i al ef f ect s of t hese f ai l ur es on t h e syst em per f or mance and t hei r
r el at i ve sever i t y i n t er ms of saf et y, ext end of damage, and i mpact on
mi ssi on success. FM ECA i s per f or med t o i dent i f y r el i abi l i t y, mai nt enance,
saf et y and suppor t abi l i t y pr obl ems r esul t i ng f r o m t he ef f ect s of a
pr oduct / pr ocess f ai l ur e. It i s an excel l ent met hodol ogy f or i dent i f yi ng and
i nvest i gat i ng pot ent i al pr oduct weakn esses. FM ECA est abl i shes a det ai l ed
st udy of t he pr oduct d esi gn, manuf act ur i ng oper at i on or di st r i but i on t o
det er mi ne whi ch f eat ur es ar e cr i t i cal t o var i ous modes of f ai l ur e. The
FM ECA con cept w as devel oped by US def ence i ndust r i es i n t h e 1950s, t o
i mpr ove t h e r el i abi l i t y of mi l i t ar y equi pment . Si nce t h en, FM ECA has
become an i mpor t ant t ool s appl i ed by al m ost al l i ndust r i es ar ound t he
wor l d t o i mpr ove t he r el i abi l i t y, mai nt ai nabi l i t y and suppor t abi l i t y of t hei r
pr oduct . It i s cl ai med t hat a mor e r i gor ous FM ECA anal ysi s woul d have
avoi ded t h e di sast r ous expl osi on of t he Chal l enger l aunch on 28
t h
Januar y
1986.
The t hr ee pr i nci p al st udy ar eas i n FM ECA anal ysi s ar e t he f ai l ur e mod e,
f ai l ur e ef f ect and f ai l ur e cr i t i cal i t y. Fai l ur e mod e an al ysi s l i st s al l possi bl e
mod e t he f ai l ur e woul d occur whi ch i ncl ude t he condi t i on, t he compon ent s
i nvol ved, l ocat i on et c. The f ai l ur e ef f ect anal ysi s i ncl udes t h e st udy of t he
l i kel y i mpact of f ai l ur e on t he per f or mance of t he w hol e pr oduct and t he
pr ocess. The cr i t i cal i t y anal ysi s exami nes how cr i t i cal a f ai l ur e woul d be f or
t he o per at ion and saf e use of t he pr oduct . The cr i t ical i t y mi ght r ange f r om
mi nor f ai l ur e t hr ou gh l ower i ng of per f or mance, shut down of t h e pr oduct ,
saf et y and envi r onment al hazar d t o a cat ast r ophi c f ai l ur e. Thi s anal ysi s i s
best ut i l i sed dur i ng t h e ear l y desi gn and d evel opment phase of new
syst ems, and i n t he eval uat i on of exi st i ng syst em (D Ver ma, 1993).
The act ual FM ECA per f or med coul d b e bot h quant i t at i ve and qual i t at i ve
based on t he i nf or mat i on avai l abl e t o t he anal yst . In put r equi r ement s f or
FM ECA anal ysi s i ncl ude r el i abi l i t y dat a, t hei r modes of f ai l ur e, and t he
est i mat ed cr i t i cal i t y of t he f ai l ur es. Addi t i onall y, t he pr obabi l i t i es of
det ect i on f or t he var i ous f ai l ur e mod es ar e al so r equi r ed. A pr er equi si t e f or
t he su ccessf ul compl et i on of FM ECA i s good knowl ed ge of , and f ami l i ar i t y
w i t h t he pr oduct / pr ocess bei ng anal ysed and i t s desi gn and f unct i onal i t y ( D
Ver ma, 1993).
45.1 Procedural Steps in the FMECA analysis
The pr ocedur al st eps i n FM ECA anal ysi s dep end t o a cer t ai n ext end what
pr oduct or pr ocess i s bei ng exami ned. The sequ ence of st eps f ol l owed t o
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 201
accompl i sh t he f ai l ur e modes, ef f ect and cr i t i cal i t y anal ysi s i s depi ct ed i n
Fi gur e 11.1. The f ol l owi ng ar e t he key st eps i nvol ved i n t he FM ECA anal ysi s:
1. Ident i f i cat i on of t he syst em r equi r ement s, by def i ni ng t he basi c
r equi r ement s f or t he syst em i n t er ms of i nput cr i t er i a f or desi gn.
Dur i ng t he syst em r equi r ement def i ni t i on, t he f ol l owi ng t asks
shoul d be addr essed (Ref er t o Bl anchar d and Fabr ycky 1999 f or
det ai l ed di scussi on).
What is expected from the system in terms of operation and
performance.
What is the customer requirements with respect to reliability,
maintainability and supportability
How the system is used in terms of hours of operation/number of
cycles per day etc.
What are the requirements for disposal after the system is
withdrawn from service.
2. Accompl i sh f unct i onal anal ysi s (Funct i onal anal ysi s i s a syst emat i c
appr oach t o syst em d esi gn and devel opment , whi ch empl oys
f unct i onal appr oach as a basi s f or i dent i f i cat i on of desi gn
r equi r ement s f or each hi er ar chi cal l evel of t he syst em. Funct i onal
anal ysi s i s accompl i shed t hr ough f unct i onal f l ow di agr am t hat
por t r ays t he syst em d esi gn r equi r ement s i l l ust r at i ng ser i es and
par al l el r el at i onshi ps and f unct i onal i nt er f aces).
3. Accompl i sh r equi r ement s al l ocat i on, t hat i s f or a speci f i ed
r equi r ement at syst em l evel , w hat shoul d be speci f i ed at uni t and
assembl y l evel . Syst em ef f ect i veness f act or s such as r el i abi l i t y,
mai nt ai nabi l i t y and suppor t abi l i t y speci f i ed at syst em l evel ar e
al l ocat ed t o uni t and assembl y l evel .
4. Ident i f i cat i on of al l possi bl e f ai l ur e modes f or t he syst em as wel l as
t he subsyst em, modul es and compon ent s.
5. Det er mi ne cause of f ai l ur es, wh i ch coul d b e desi gn and
manuf act ur i ng d ef i ci ency, agei ng and wear  out , acci dent al damage,
t r anspor t at i on and handl i ng, mai nt enance i nduced f ai l ur es.
6. Ident i f y t h e ef f ect s of f ai l ur e. Ef f ect of f ai l ur e mi ght r ange f r om
cat ast r ophi c f ai l ur e t o mi nor per f or mance degr adat i on.
7. Assess t he pr obabi l i t y of f ai l ur e. Thi s can be achi eved by anal ysi ng
t he f ai l ur e dat a and i dent i f yi ng t he t i met of ai l ur e di st r i but i on.
8. Ident if y t he cr i t i cal i t y of f ai l ur e. Fai l ur e cr i t i cali t y can be classi f i ed i n
any one of f our cat egor i es, depen di ng upon t h e f ai l ur e ef f ect s as
f ol l ow s
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 202
a) Minor failure Any failure that doesn`t have any noticeable
affect on the performance of the system.
b) Major failure Any failure that will degrade the system
performance beyond an acceptable limit.
c) Critical failure Any failure that would affect safety and
degrade the system beyond an acceptable limit.
d) Catastrophic failure Any failure that could result in significant
system damage and may cause damage to property, serious
injury or death.
9. Compute the Risk Priority Number (RPN) by multiplying the
probability of failure, the severity of the effects and the likelihood of
detecting a failure mode.
10. Initiate corrective action that will minimise the probability of failure
or effect of failure that show high RPN.
Figure 11.1 Sequence of steps involved in FMECA
Define System
Requirements
Accomplish
Functional Analysis
Requirements
Allocation
Identify Failure
Modes
Determine Causes of
Failure
Determine Effects of
Failure
Assess the
probability of Failure
Assess the
probability of Failure
Assess the Criticality
of Failure
Analyse failure mode
Criticality (PRN)
Feedback and corrective action
loop
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 203
45.2 Risk Priority Number
Ri sk Pr i or i t y Number s pl ay a cr uci al r ol e i n sel ect i ng t he most si gni f i cant
i t em t hat wi l l mi ni mi se t he f ai l ur e or ef f ect of f ai l ur e. As ment i oned ear l i er ,
RPN i s cal cul at ed by mul t i pl yi ng t he pr obabi l i t y of f ai l ur e, t he sever i t y of
t he ef f ect s of f ai l ur e and l i kel i hood of f ai l ur e det ect i on. That i s:
RPN = FP × FS × FD (11.1)
Wh er e, FP i s t he Fai l ur e pr obabi l i t y, FS i s t he f ai l ur e sever i t y and FD
denot es t h e f ai l ur e det ect i on pr obabi l i t y. Tabl es 11.1 – 11.3 gi ves possi bl e
r at i ngs f or pr obabi l i t y of f ai l ur e, sever i t y of f ai l ur e and f ai l ur e det ect i on.
Not e t hat , t he r at i ngs gi ven i n t h e t abl es 11.111. 3 ar e onl y su ggest ed
r at i ngs.
Table 11.3. Rating scales for occurrence of failure
Description Rating
Remote probability of occurrence 1
Low probability of occurrence 2  3
Moderate probability of occurrence 4  6
High probability of occurrence 7 8
Very High probability of occurrence 9  10
Table 11.2 Rating scales for severity of failure
Description Rating
Minor failure 1 2
Major Failure 3 5
Critical Failure 6 9
Catastrophic Failure 10
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 204
Description Rating
Table 11.3. Rating scales for detection of failure
Description Rating Probability of Detection
Remote probability of
detection
1 0 0.05
Low probability of
detection
2  3 0.06 0.15
Moderate probability of
detection
4  5 0.16 0.35
High probability of
detection
6  8 0.36 0.75
Very high probability of
detection
9  10 0.76 1.00
Assume t hat a f ai l ur e mod e has f ol l ow i ng r at i ngs f or pr obabi l i t y of f ai l ur e,
f ai l ur e sever i t y and f ai l ur e det ect i on:
Fai l ur e pr obabi l i t y = 7
Fai l ur e sever i t y = 4
Fai l ur e det ect i on = 5
Then t he r i sk pr i or i t y number f or t hi s par t i cul ar f ai l ur e mode i s gi ven by 7 ×
4 × 5 = 140. Ri sk pr i or i t y nu mb er f or al l t he f ai l ur e modes ar e cal cul at ed
and pr i or i t y i s gi ven t o t he one wi t h hi gh est RPN f or el i mi nat i ng t he f ai l ur e.
Thi s i s usual l y achi eved usi ng Par et o anal ysi s wi t h a f ocus on f ai l ur e mod e,
f ai l ur e cause an d f ai l ur e cr i t i cal i t y. Out put s f r o m a pr oper l y conduct ed
FM ECA can be used i n d evel opi ng a cost ef f ect i ve mai nt enan ce anal ysi s,
syst em saf et y hazar d anal ysi s, and l ogi st i c suppor t anal ysi s.
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 205
46. FAULT TREE ANALYSIS (FTA)
Faul t t r ee anal ysi s i s a deduct i ve appr oach i nvol vi ng gr aphi cal enumer at i on
and anal ysi s of t he di f f er ent w ays i n w hi ch a par t i cul ar syst em f ai l ur e can
occur , and t he pr obabi l i t y of i t s occur r en ce. It st ar t s w i t h a t opl evel event
(f ai l ur e) and wor ks backwar d t o i den t i f y al l t he possi bl e causes and
t her ef or e t he or i gi ns of t hat f ai l ur e. Dur ing t he ver y ear l y st ages of syst em
desi gn pr o cess, and i n t he absence of i nf or mat i on r equi r ed t o compl et e a
FM ECA, f aul t t r ee anal ysi s (FTA) i s of t en conduct ed t o gai n i nsi ght i nt o
cr i t i cal aspect s of sel ect ed desi gn concep t s. Usual l y, a separ at e f aul t t r ee i s
devel oped f or ever y cr i t i cal f ai l ur e mod e or undesi r ed TopLevel event .
At t ent i on i s f ocused on t hi s t opl evel event and t he f i r st t i er causes
associ at ed wi t h i t . Each f i r st t i er cause i s next i nvest i gat ed f or i t s causes,
and t hi s pr ocess i s cont i nued. Thi s ‘ TopDow n’ cau sal hi er ar chy and t he
associ at ed pr obabi l i t i es, i s cal l ed a Faul t Tr ee.
One of t he out pu t s f r o m a f aul t t r ee anal ysi s i s t h e pr obabi l i t y of occur r ence
of t he t opl evel even t or f ai l ur e. If t hi s pr obabi l i t y i s unaccept abl e, f aul t
t r ee anal ysi s pr ovi des t h e d esi gn er s wi t h an i nsi ght i nt o aspect s of t he
syst em t o whi ch r edesi gn can be di r ect ed or compensat or y pr ovi si ons be
pr ovi ded such as r edundancy. Th e FTA can have most i mpact i f i ni t i at ed
dur i ng t h e concept ual and pr el i mi nar y desi gn phase wh en desi gn and
conf i gur at i on changes can be most easi l y and cost ef f ect i vel y i mpl emen t ed.
The l ogi c used i n devel opi ng and anal ysi ng a f aul t t r ee has i t s f oundat i ons in
Bool ean Al gebr a. The f ol l owi ng st eps ar e used t o car r y out FTA (Fi gur e
11.2).
1. Identify the toplevel event The most important step is to identify and
define the toplevel event. It is necessary to specific in defining the top
level event, a generic and nonspecific definition is likely to result in a
broad based fault tree which might be lacking in focus.
2. Develop the initial fault tree Once the toplevel event has been
satisfactorily identified, the next step is to construct the initial causal
hierarchy in the form of a fault tree. Techniques such as Ishikawas
cause and effect diagram can prove beneficial. While developing the
fault tree all hidden failures must be considered and incorporated. For the
sake of consistency, a standard symbol is used to develop fault trees.
Table 11.4 depicts the symbols used to represent the causal hierarchy and
interconnects associated with a particular toplevel event. While
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 206
constructing a fault tree it is important to break every branch down to a
reasonable and consistent level of detail.
3. Analyse the Fault Tree The third step in FTA is to analyse the initial
fault tree developed. The important steps in completing the analysis of a
fault tree are 1. Delineate the minimum cutsets, 2. Determine the
reliability of the toplevel event and 3. Review analysis output.
Figure 11.2 Steps involved in a fault tree analysis.
Table 11.4. Fault tree construction symbols
Symbol Description
The Ellipse represents the toplevel event (thus always appears at the
very top of the fault tree).
The rectangle represents an intermediate fault event. A rectangle can
appear anywhere in a fault tree except at the lowest level in the
hierarchy.
A circle represents the lowest level failure event, also called a basic
event.
The diamond represents an undeveloped event, which can be further
broken. Very often, undeveloped events have a substantial amount of
complexity below and can be analysed through a separate fault tree.
This symbol represents the AND logic gate. In this case, the output is
realised only after all the associated inputs have been received.
Identify Top
Level
Event
Develop the
Initial
Fault Tree
Analyse the
Fault
Tree
Delineate the
Minimal Cut
sets
Determine Top
Level Event
Reliability
Review Analysis
Output
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 207
Symbol Description
This symbol represents the OR logic gate. In this case, any one or
more of the inputs need to be received for the output to be realised.
47. FAULT TREE ANALYSIS CASE STUDY
PASSENGER ELEVATOR
In t hi s sect i on w e di scuss a case st udy on f aul t t r ee anal ysi s of a passenger
el evat or (M ai n sour ce, D Ver ma, 1993). Consi der a passenger el evat or
depi ct ed i n Fi gur e 11.3. We consi der t wo maj or assembl i es f or FTA 1.
Cont r ol assembl y and 2. Dr i ve/ suspensi on assembl y. Al l dr i ve assembl y
f ai l ur es ar e gener al i sed as ‘ mot or f ai l ur es’ and ‘ ot her f ai l ur es’ whi l e cont r ol
uni t f ai l ur es ar e gener al i sed as ‘ har dwar e f ai l ur es’ and ‘ sof t w ar e f ai l ur es’
f or t he sake of si mpl i ci t y.
Fi gur e 11.3 Schemat i c di agr am of a passenger el evat or
The cont r ol assembl y consi st s of a mi cr opr ocessor , whi ch awai t s an
oper at or si gnal r equest t o move t he car t o a cer t ai n l evel . The con t r ol uni t
act i vat es dr i ve uni t t hat moves t h e car t o t hat l evel and opens t he el evat or
Passenger
Car
Shaft
Level 3
Level 2
Level 1
Level 0
Drive Unit
Control
Unit
Power
Supply
Weight
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 208
door once t he car comes t o a st op. Swi t ches exi st at each l evel and i nsi de
t he car al l o w i ng t he cont r ol l er t o know w her e t he car i s at any t i me.
Dr i ve/ susp ensi on assembl y hol ds t h e car susp ended w i t hi n t he shaf t and
moves i t t o t he cor r ect l evel as i ndi cat ed by t h e cont r ol uni t . The Dr i ve uni t
moves or st ops t h e car onl y when pr ompt ed t o do so by t he cont r ol uni t .
The br ake uni t i s desi gned t o hod t h e car st at i onar y when power i s
r emoved and t o al l ow t he mot or shaf t t o t ur n when power i s appl i ed.
We d ef i ne t h e t opl evel event i n t hi s case i s ‘ passenger i nj ur y occur s’ . The
f ol l ow i ng ar e t he possi bl e syst em oper at i ng condi t i ons:
A. El evat or oper at i ng pr oper l y.
B. Car st ops bet w een l evel s.
C. Car f al l s f r eel y.
D. Car ent r y door opens i n t he absence of car .
In t hi s case, oper at i ng condi t i ons ‘ C’ and ‘ D’ ar e of concer n. The i ni t i al
f aul t t r ee i s shown i n Fi gur e 11.4.
Fi gur e 11.4 Ini t i al f aul t Tr ee
In Fi gur e 11.4, G1 i s r epr esent s t he OR l ogi c gat e and t he event s 1, 2 and
3 ar e as def i ned bel ow :
Event 1 – Passenger i nj ur y occur s
Event 2 – Car f r ee f al l s
Event 3 – Door opens w i t hout car pr esent .
Thus, t he t opl evel event (passenger i nj ur y occur s) can be ei t her due t o
car f r ee f al l or door opens wi t hout t he car pr esent .
1
2 3
G1
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 209
Now t he event , car f r ee f al l , can f ur t her anal ysed by t r eat i ng i t as a t op
l evel event , r esul t i ng i n a f aul t t r ee depi ct ed i n Fi gur e 11.5. In Fi gur e
11.5, G2 i s agai n a OR gat e and t he event s 4, 5 and 6 ar e def i ned bel ow :
Event 4 – Cabl e sl i ps of pul l ey
Event 5 – Hol di ng br ake f ai l ur e
Event 6 – Br oken cabl e
Event 4 and 6 ar e undevel oped event , whi ch can f ur t her br oken, w hi ch
can be f ur t her anal ysed usi n g a separ at e f aul t t r ee. Event 5 i s an
i nt er medi at e event .
Fi gur e 11.5 Fur t her FTA anal ysi s of t he event car f r ee f al l
2
G2
4 6 5
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 210
Fi gur e 11.6 Faul t t r ee f or t he event , t he door opens er r oneousl y
The event 3 can be f ur t h er anal ysed t o f i nd t he causes, Fi gur e 11.6 depi ct s
FTA f or t he event 3, door opens wi t hout t he car pr esent . Thi s can be caused
due t o t he f ol l owi ng event s:
Event 7 – Door cl ose f ai l ur e
Event 8 – Car not at l evel
Event 9 – Lat ch f ai l ur e
For t he event , door opens er r oneousl y, t o occur , event s 7 and 8 must
happen, t hus w e have a AND gat e G3. The door cl ose f ai l ur e can be caused
ei t her due t o t he l at ch f ai l ur e or du e t o cont r ol l er er r or (deno t ed by OR
gat e, G4). Combi ni ng f aul t t r ees depi ct ed i n Fi gur es 11.411.6, we can
const r uct a compl et e (al most ) f or t he event , passen ger i nj ur y, as shown i n
Fi gur e 11.7. Not e t hat event s 4, 5, 6, 8 and 9 can be f ur t her expanded t o
f i nd t he causes usi ng f aul t t r ee anal ysi s. Th e pr obabi l i t y f or t he occur r ence
of t he t opl evel event can be cal cul at ed o nce t he t i met o f ai l ur e and
pr obabi l i t y of occur r ence of al l t h e event s ar e known. If t he der i ved t op
l evel pr obabi l i t y i s unaccept abl e, necessar y r edesi gn or compensat i on
ef f or t s shoul d be i dent i f i ed and i ni t i at ed. As i t i s a si mpl e mat hemat i cal
cal cul at i on, i t i s not cover ed i n t hi s book.
3
G4
7 8
G5
9
Controller
Error
11. Desi gn f or Rel i abi l i t y, M ai nt enance and Logi st i c Suppor t 211
Fi gur e 11.7 Faul t t r ee f or t he event , passenger i nj ur y
1
2 3
G1
4 6 5
G2 G3
7 8
G4
9
Controller
Error
Chapter 12
Analysis of Reliability, Maintenance and
Supportability Data
Of t en st at i st i cs ar e used as a dr unken man uses l amp post s… f or suppor t
r at her t han i l l umi nat i on.
To pr edi ct var i ous r el i abi l i t y char act er i st i cs of an i t em, as w el l as i t s mai nt ai nabi l i t y and suppor t abi l i t y
f unct i on, i t i s essent i al t hat w e have suf f i ci ent i nf or mat i on on t he t i me t o f ai l ur e, t i me t o r epai r
(mai nt ai n) and t i me t o suppor t char act er i st i cs of t hat i t em. In most cases t hese char act er i st i cs ar e
expr essed usi ng t heor et i cal pr obabi l i t y di st r i but i ons. Thus, t he pr obl em w hi ch ever y l o gi st i ci an f ace i s
t he sel ect i on of t he appr opr i at e di st r i but i on f unct i on t o descr i be t he empi r i cal dat a (obt ai ned f r om dat a
capt ur i ng sour ces) usi ng t heor et i cal pr obabi l i t y di st r i but i ons. Once t he di st r i but i on i s i dent i f i ed, t hen
one can ext r act i nf or mat i on about t he t ype of t he hazar d f unct i on and ot her r el i abi l i t y char act er i st i cs
such as mean t i me bet w een f ai l ur es and f ai l ur e r at e et c. In t he case of mai nt enance and suppor t abi l i t y
dat a, w e w oul d i d ent i f y t he mai nt ai nabi l i t y and suppor t abi l i t y f unct i on as i n t he case of r el i abi l i t y dat a
and t hen comput e M TTR and M TTS.
To st ar t w i t h w e l ook at w ays of f i t t i ng pr obabi l i t y di st r i but i ons t o i nser vi ce dat a, t hat i s t he dat a
r el at i ng t o t he age of t he component s at t he t i me t hey f ai l ed w hi l e t hey w er e i n oper at i on (i n
mai nt enance and l ogi st i c suppor t we anal yse t he dat a cor r espondi ng t o t he mai nt enance and suppor t
t ask compl et i on t i mes). We l ook at t hr ee popul ar t ool s; 1. Pr obabi l i t y paper s, 2. Li near r egr essi on, and
3. M axi mum l i kel i hood est i mat es t o i dent i f y t he best di st r i but i on usi ng w hi ch t he dat a can be expr essed
and t o est i mat e t he cor r espondi ng par amet er s of t he di st r i but i on. In t he sect i on on “ censor ed dat a” w e
r ecogni se t hat ver y of t en w e do not have a compl et e set of f ai l ur e dat a. We may wi sh t o det er mi n e
w het her a new ver si on of a component i s mor e r el i abl e t han a pr evi ous ver si on t o deci de w het her w e
have cur ed t he pr obl em (of pr emat ur e f ai l ur es, say). Of t en, compon ent s wi l l be r epl aced bef or e t hey
have act ual l y f ai l ed, possi bl y because t hey have st ar t ed t o cr ack, t hey have been damaged or t hey ar e
show i ng si gns of excessi ve w ear . We may have a number of syst ems und er goi ng t est i ng t o det er mi n e
w het her t he pr oduct i s l i kel y t o meet t he var i ous r equi r ement s but w e need t o go i nt o pr oduct i on
bef or e t hey have al l f ai l ed. Ther e i s usef ul dat a t o b e gl eaned f r om t he ones t hat have not f ai l ed as w el l
as f r om t he ones t hat have f ai l ed. If a component i s bei ng used i n a number of di f f er ent syst ems, i t may
be r easonabl e t o assume t hat t he f ai l ur e mechani sm i n each of t hese i nst ances w i l l be si mil ar . Even
t hough t he way t he di f f er ent syst ems oper at e may be di f f er ent , i t i s st i l l l i kel y t hat t he shap e of t h e
f ai l ur e di st r i but i on wi l l be same and t hat onl y t he scal e wi l l be di f f er ent .
Even r el at i vel y si mpl e syst ems can f ai l i n a number of di f f er ent ways and f or a number of di f f er ent
r easons. Suppose w e w i sh t o f ast en t w o pi eces of met al t oget her usi ng a nut and bol t . If w e over 
t i ght en t he nut , w e mi ght st r i p t he t hr ead or w e mi ght shear t he bol t . If w e do not put t he nut on
squar el y, we coul d cr oss t he t hr eads and hence w eaken t he j oi nt . If t he t wo pi eces of met al ar e bei ng
f or ced apar t t hen t he st r ess on t he nut and bol t may cause t he t hr ead t o st r i p ei t her i nsi de t he nut or on
t he out si de of t he bol t or i t may cause t he bol t t o exceed i t s el ast i c and pl ast i c l i mi t s unt i l i t event ual l y
213
br eaks. If t he j oi nt i s subj ect t o excessi ve heat t hi s coul d accel er at e t he pr ocess. Equal l y, i f i t i s i n ver y
l ow t emper at ur es t hen t he bol t i s l i kel y t o become mor e br i t t l e and br eak under l ess st r ess t han at
nor mal t emper at ur es. If t he di amet er of t he bol t i s t ow ar ds t he l ow er l i mi t of i t s t ol er ance and t he
i nt er nal di amet er of t he nut i s t ow ar ds t he upper l i mi t t hen t he amount of met al i n cont act may not be
suf f i ci ent t o t ake t he st r ai ns i mposed. As t he t w o component s age, cor r osi on may cause t he amount of
met al i n cont act t o be even f ur t her r educed. It may al so change t he t ensi l e st r engt h of t he met al s and
cause pr emat ur e f ai l ur e.
Component s may t her ef or e f ai l due t o a number of f ai l ur e modes. Each of t hese modes may be mor e or
l ess r el at ed t o t he age. One w oul d not expect cor r osi on t o be t he cause of f ai l ur e dur i ng t he ear l y
st ages of t he component ’ s l i f e, unl ess i t w as subj ect ed t o except i onal l y cor r osi ve chemi cal s. On t he
ot her hand, i f t he component s have been badl y made t hen one mi ght expect t o see t hem f ai l ver y soo n
af t er t he uni t has been assembl ed.
Ver y of t en, a possi bl y smal l , number of component s may f ai l unexp ect edl y ear l y. On f ur t h er
i nvest i gat i on i t may be f ound t hat t hey w er e al l made at t he same t i me, f r om t he same i ngot of met al or
by a par t i cul ar suppl i er . Such a phenomenon i s commonl y r ef er r ed t o as a bat chi ng pr obl em.
Unf or t unat el y, i n pr act i ce, al t hough i t may be possi bl e t o r ecogni se i t s pr esence, i t may not al ways be
possi bl e t o t r ace i t s or i gi n or , mor e poi gnant l y, t he ot her member s of t he same bat ch or , i ndeed, how
many t her e may be.
In deci di ng w het her a new ver si on of a component i s mor e r el i abl e t han t he ol d one, w e need t o
det er mi ne how conf i dent w e ar e t hat t he t w o di st r i but i ons ar e di f f er ent . If t hey bot h have t he same (or
near l y t he same) shapes t hen i t i s a r el at i vel y st r ai ght f or war d t ask t o det er mi ne i f t hei r scal es ar e
di f f er ent . In some cases, t he pr i mar y cause of f ai l ur e of t he or i gi n ver si on may have been el i mi nat ed or ,
at l east , si gni f i cant l y i mpr o ved but , anot her , hi t her t o r ar el y seen cause, may have become el evat ed i n
si gni f i cance. Thi s new pr i mar y cause may have a di st i nct l y di f f er ent shape t han t h e f i r st one t hat of t en
makes i t ver y di f f i cul t t o deci de bet w een t he t w o.
In t hi s chapt er , w e f i r st l ook at t he empi r i cal appr oaches f or f i ndi ng est i mat es f or M TTF, M TTR and
M TTS as w el l as f ai l ur e f unct i on, mai nt ai nabi l i t y and suppor t abi l i t y f unct i ons. Rest of t he chapt er
descr i bes some of t he w el l know n met hods f or sel ect i on of t he most r el evant t heor et i cal di st r i but i on
f unct i ons f or t he r andom var i abl es under consi der at i on.
12.48. RELIABILITY, MAINTENANCE AND SUPPORTABILITY DATA
A ver y common pr obl em i n r el i abi l i t y engi n eer i ng i s t he avai l abi l i t y of f ai l ur e dat a. In many cases get t i n g
suf f i ci ent dat a f or ext r act i n g r el i abl e i nf or mat i on i s t he most di f f i cul t t ask. Thi s may be due t he f act t hat
t her e i s no good pr o cedur e empl oyed by t he oper at or (or suppl i er ) t o col l ect t he dat a or t he i t em may
be hi ghl y r el i abl e and t he f ai l ur e i s ver y r ar e. How ever , even w i t hout any dat a, one shoul d be abl e t o
pr edi ct t he t i met of ai l ur e di st r i but i on i f not t he par amet er s. For exampl e, i f t he f ai l ur e mechani sm i s
cor r osi on, t hen i t cannot be an exponent i al di st r i but i on. Si mi l ar l y i f t he f ai l ur e cause i s ‘ f or ei gn obj ect
damage’ t hen t he onl y di st r i but i on t hat can be used i s exponent i al . The mai n pr obl em wi t h i nsuf f i ci ent
f ai l ur e dat a i s get t i ng an accur at e est i mat e f or t he sh ape par amet er . For t unat el y, we don’ t have su ch
pr obl ems wi t h mai nt enance and suppor t abi l i t y dat a. These ar e easi l y avai l abl e f r om t he peopl e who
mai nt ai n and suppor t t he i t em. The r el i abi l i t y dat a can be obt ai ned f r om t he f ol l ow i ng sour ces:
1. Fi el d dat a and t he i nser vi ce dat a f r om t he oper at or usi ng st andar d dat a capt ur i ng t echni ques.
Ther e ar e st andar d f ai l ur e r epor t i ng f or ms f or t he pur pose of capt ur i ng desi r ed i nf or mat i on
r egar di ng t he r el i abi l i t y of t he i t em under consi der at i on. Unf or t unat el y, al l t hese f or ms ar e
214
f l aw ed, as t hey r ecor d onl y M TBF (or M TTR and M TTS i n case of mai nt enance and suppor t ).
Just t he val u e of M TBF al one may not be enough f or many anal yses concer ni ng r el i abi l i t y
(si mi l arl y, i n t he case of mai nt enance (suppor t ), i nf or mat i on on M TTR (M TTS) i s not enough f or
compl et e anal yses).
2. Fr om l i f e t est i ng t hat i nvol ves t est i ng a r epr esent at i ve sampl e of t he i t em under cont r ol l ed
condi t i ons i n a l abor at or y t o r ecor d t he r equi r ed dat a. Somet i mes, t hi s mi ght i nvol ve
‘ accel er at ed l i f e t est i ng’ (ALT) and ‘ hi ghl y accel er at ed l i f e t est i ng’ (HALT) dependi ng on t h e
i nf or mat i on r equi r ed.
As ment i oned ear l i er , i n some cases i t i s not possi bl e t o get a compl et e f ai l ur e dat a f r om a sampl e. Thi s
i s because some of t he i t ems may not f ai l duri ng t he l i f e t est i ng (al so i n t he i nser vi ce dat a). These t ypes
of dat a ar e cal l ed ‘ censor ed dat a’ . If t he l i f e t est i ng exp er i ment i s st opped bef or e al l t he i t ems have
f ai l ed, i n w hi ch cases onl y t he l ow er bound i s know n f or t he i t ems t hat have not f ai l ed. Such t ype of
dat a i s know n as ‘ r i ght censor ed dat a’ . In f ew cases onl y t he upper bound of t he f ai l ur e t i me may be
know n, such t ype of dat a i s cal l ed ‘ l ef t censor ed dat a’ .
12.49. ESTIMATION OF PARAMETERS  EMPIRICAL APPROACH
The obj ect i ve of empi r i cal met hod i s t o est i mat e f ai l ur e f unct i on, r el i abi l i t y f unct i on, hazar d f unct i on,
M TTF (or M TTR and M TTS) f r om t he f ai l ur e t i mes (or r epai r and suppor t t i mes). Empi r i cal appr oach i s
of t en r ef er r ed as nonpar amet r i c appr oach or di st r i but i on f r ee appr oach. In t he f ol l ow i ng sect i ons w e
di scuss met hods f or est i mat i ng var i ous per f or mance measur es used i n r el i abi l i t y, mai nt enance and
suppor t f r om di f f er ent t ypes of dat a.
Estimation of Performance Measures  Complete Ungrouped Data
Compl et e ungr ouped dat a r ef er s t o a r aw dat a (f ai l ur e, r epai r or suppor t ) w i t hout any censor ed dat a.
That i s, t he f ai l ur e t i mes of t he whol e sampl e under consi der at i on ar e avai l abl e. For exampl e, l et t
1
, t
2
,
…, t
n
, r epr esent s n or der ed f ai l ur e t i mes such t hat t
i
≤ t
i+1
. Then t he possi bl e est i mat e f or f ai l ur e
f unct i on (cumul at i ve f ai l ur e di st r i but i on at t i me t
i
) i s gi ven by:
n
i
t F · ) (
^
(12.1)
A t ot al of i uni t s f ai l by t i me t out of t he t ot al n i n t h e sampl e. Thi s w i l l make F(t
n
) = n / n = 1. That i s,
t her e i s a zer o pr obabi l i t y f or any i t em t o sur vi ve beyond t i me t
n
. Thi s i s ver y unl i kel y, as t he t i mes ar e
dr awn f r om a sampl e and i t i s ext r emel y unl i kel y t hat any sampl e woul d i ncl ude t he l on gest sur vi val
t i me. Thus t he equat i on (12.1) under est i mat es t he component sur vi val f unct i on. A number of
mat hemat i ci ans have t r i ed t o f i nd a sui t abl e al t er nat i ve met hod of est i mat i ng t he cumul at i ve f ai l ur e
pr obabi l i t y. These r ange f r om usi ng n+1 i n t he denomi nat or t o usi ng 0.5 i n t he numer at or and +0.5 i n
t he denomi nat or . The one t hat gi ves t he best appr oxi mat i on i s based on medi an r ank. Ber nar d' s
appr oxi mat i on t o t he medi an r ank appr oach f or cumul at i ve f ai l ur e pr obabi l i t y i s gi ven by
4 . 0
3 . 0
) (
^
+
−
·
n
i
t F
i
(12.2)
215
Thr oughout t hi s chapt er w e use t he above appr oxi mat i on t o est i mat e t he cumul at i ve f ai l ur e di st r i but i on
or f ail ur e f unct i on. Fr om equat i on (12.2), t he est i mat e f or r el i abi l i t y f unct i on can be obt ai ned as
4 . 0
7 . 0
4 . 0
3 . 0
1 ) ( 1 ) (
^ ^
+
+ −
·
+
−
− · − ·
n
i n
n
i
t F t R
i i
(12.3)
The est i mat e f or t he f ai l ur e densi t y f unct i on f (t ) can be obt ai ned usi ng
1
1
1
^ ^
^
,
) ( ) (
) (
+
+
+
≤ ≤
−
−
·
i i
i i
i i
t t t
t t
t F t F
t f (12.4)
Est i mat e f or t he hazar d f unct i on can be obt ai ned by usi ng t he r el at i on bet w een t he r el i abi l i t y f unct i on
R(t ) and t he f ai l ur e densi t y f unct i on f (t ). Ther ef or e,
1
^ ^ ^
for ) ( ) ( ) (
+
< < ·
i i
t t t t R t f t h (12.5)
An est i mat e f or t he mean t i me t o f ai l ur e (or mean t i me t o r epai r or mean t i me t o suppor t ) can be
di r ect l y obt ai ned f r om t he sampl e mean. That i s,
∑
·
·
n
i
i
n
t
MTTF
1
^
(12.6)
Est i mat e f or t he var i ance of t he f ai l ur e di st r i but i on can be obt ai ned f r om t he sampl e var i ance, t hat i s
∑
·
−
−
·
n
i
i
n
MTTF t
s
1
2
^
2
1
) (
(12.7)
Est i mat e f or M TTR (M TTS) and Var i ance of t i me t o r epai r di st r i but i on (t i me t o suppor t di st r i but i on) can
be obt ai ned by r epl aci ng f ai l ur e t i mes by r epai r t i mes (suppor t t i mes) i n equat i on (12.6) and (12.7)
r espect i vel y.
Confidence Interval
It i s al w ays of t he i nt er est t o know t he r ange i n w hich t he measur es such as M TTF, M TTR and M TTS
mi ght l i e wi t h cer t ai n conf i d ence. The r esul t i ng i nt er val i s cal l ed a conf i dence i nt er val and t he
pr obabi l i t y t hat i t cont ai ns t he est i mat ed par amet er i s cal l ed i t s conf i dence l evel or conf i dence
coef f i ci ent . For exampl e, i f a conf i dence i nt er val has a conf i dence coef f i ci ent equal t o 0.95, w e cal l i t a
95% conf i dence i nt er val .
To der i ve a (1α) 100% conf i dence i nt er val f or a l ar ge sampl e we use t he f ol l owi ng expr essi on:
) (
2 /
^
n
z MTTF
σ
α
t (12.8)
216
Wher e z
α/ 2
i s t he z val u e (st andar d nor mal st at i st i c) t hat l ocat es an ar ea of α/ 2 t o i t s r i ght and can be
f ound f r om t he nor mal t abl e. σ i s t he st andar d devi at i on of t he popul at i on f r om w hi ch t he popul at i on
w as sel ect ed and n i s t he sampl e si ze. The above f or mul a i s val i d w henever t he sampl e si ze n i s gr eat er
t han or equal t o 30. The 90%, 95% and 99% conf i dence i nt er val f or M TTF w i t h sampl e si ze n ≥ 30 ar e
gi ven bel ow:
,
_
¸
¸
× t
n
MTTF
σ
645 . 1 confidence % 90
^
(12.9)
,
_
¸
¸
× t
n
MTTF
σ
96 . 1 confidence % 95
^
(12.10)
,
_
¸
¸
× t
n
MTTF
σ
58 .. 2 confidence % 99
^
(12.11)
When t he number of dat a i s smal l (t hat i s w hen n i s less t han 30), t he conf i dence i nt er val i s based on t
di st r i but i on. We use t he f ol l owi ng expr essi on t o cal cul at e (1α)100% conf i dence i nt er val .
,
_
¸
¸
t
n
s
t MTTF
2 /
^
α
(12.12)
w her e t
α/ 2
i s based on (n1) degr ees of f r eedom and can be obt ai ned f r om t di st r i but i on t abl e (r ef er
appendi x).
Exampl e 12.1
Ti me t o f ai l ur e dat a f or 20 car gear boxes of t he model M 2000 i s l i st ed i n Tabl e 12.1. Fi nd:
1. Estimate of failure function and reliability function.
2. Plot failure function and the reliability function.
3. Estimate of MTTF and 95% confidence interval.
Table 12.1. Failure data of gearboxes in miles
1022 1617 2513 3265 8445
9007 10505 11490 13086 14162
14363 15456 16736 16936 18012
19030 19365 19596 19822 20079
SOLUTION:
The f ai l ur e f unct i on and r el i abi l i t y f unct i on can be est i mat ed usi ng equat i ons 12.2 and 12.3. Tabl e 12.2
show s t he est i mat ed val u es of f ai l ur e f unct i on and r el i abi l i t y f unct i on.
217
Tabl e 12.2. Est i mat e f or f ai l ur e and r eli abi l i t y f unct i on.
Fai l ur e dat a
) (
^
i
t F ) (
^
i
t R
1022 0.0343 0.9657
1617 0.0833 0.9167
2513 0.1324 0.8676
3265 0.1814 0.8186
8445 0.2304 0.7696
9007 0.2794 0.7206
10505 0.3284 0.6716
11490 0.3774 0.6225
13086 0.4264 0.5736
14162 0.4754 0.5246
14363 0.5245 0.4755
15456 0.5735 0.4265
16736 0.6225 0.3775
16936 0.6716 0.3284
18012 0.7206 0.2794
19030 0.7696 0.2304
19365 0.8186 0.1814
19596 0.8676 0.1324
19822 0.9167 0.0833
20079 0.9657 0.0343
The failure function and the reliability function graph are shown in Figure 12.1 and 12.2 respectively.
Fi gur e 12.1 Est i mat e of f ai l ur e f unct i on f or t he dat a show n i n Tabl e 12.1
0
0.2
0.4
0.6
0.8
1
0 5000 10000 15000 20000 25000
Age of the car(in miles)
F
a
i
l
u
r
e
f
u
n
c
t
i
o
n
218
Fi gur e 12.2 Est i mat ed r el i abi l i t y f unct i on f or t he dat a gi ven i n Tabl e 12.2
The est i mat e f or mean t i me t o f ai l ur e i s gi ven by:
∑
·
· ·
20
1
^
5 . 12725
20
i
i
t
MTTF miles.
Est i mat e f or t he st andar d devi at i on i s gi ven by
miles 16 . 14827
1
) (
1
2
^
·
−
−
· ∑
·
n
i
i
n
MTTF t
s
As t he sampl e dat a i s l ess t han 30, w e use equat i on (12.12) t o f i nd t he 95% conf i dence l evel . Fr om t 
t abl e t he val ue of t
0.025
f or (n1) = 19 i s gi ven by 2.093. The 95% conf i dence l evel f or M TTF i s gi ven by:
) 19 / 16 . 14827 ( 093 . 2 5 . 12725
2 /
^
t ·
,
_
¸
¸
t
n
s
t MTTF
α
That i s, t he 95% conf i dence i nt er val f or M TTF i s (5605.98, 19845.01).
Exampl e 12.2
Ti me t aken t o compl et e r epai r t asks f or an i t em i s gi ven i n Tabl e 12.3. Fi nd t he cumul at i ve t i me t o
r epai r di st r i but i on and mean t i me t o r epai r . Fi nd 95% conf i dence l evel f or M TTR.
Tabl e 12.3. Ti me t o r epai r dat a
28 53 71 90
30 56 72 92
31 58 74 94
33 59 75 95
35 61 79 97
0
0.2
0.4
0.6
0.8
1
0 5000 10000 15000 20000 25000
Age of the car (in miles)
E
s
t
i
m
a
t
e
d
r
e
l
i
a
b
i
l
i
t
y
f
u
n
c
t
i
o
n
219
40 65 81 99
41 67 82 100
44 68 84 103
49 69 85 108
51 70 89 110
M ai nt ai nabi l i t y f unct i on can be est i mat ed usi ng f ol l owi ng expr essi on:
4 . 40
3 . 0
4 . 0
3 . 0
) (
^
−
·
+
−
·
i
n
i
t M
i
Figure 12.3 shows the estimated maintainability function.
Fi gur e 12.3.M ai nt ai nabi l i t y f unct i on f or t he dat a gi ven i n Tabl e 12.3.
M ean Ti me t o Repai r i s gi ven by:
hours 7 . 69
40
40
1
^
· ·
∑
· i
i
t
MTTR
St andar d devi at i on f or r epai r t i me i s gi ven by
hours 23.43
1
) (
1
2
^
2
·
−
−
· ∑
·
n
i
i
n
MTTR t
s
Si nce n > 30, w e use equat i on (12.10) t o cal cul at e 95% conf i dence i nt er val f or M TTR. 95% conf i dence
l evel f or M TTR i s gi ven by
) 96 . 76 , 43 . 62 (
40
43 . 23
) 96 . 1 ( 7 . 69 96 . 1
^
·
,
_
¸
¸
t ·
,
_
¸
¸
t
n
s
MTTR
0
0.2
0.4
0.6
0.8
1
0 20 40 60 80 100 120
Time
M
a
i
n
t
a
i
n
a
b
i
l
i
t
y
F
u
n
c
t
i
o
n
220
Analysis of Grouped Data
Of t en f ai l ur e dat a i s pl aced i nt o t i me i nt er val s w hen t he sampl e si ze i s l ar ge. The f ai l ur e dat a ar e
cl assi f i ed i nt o sever al i nt er val s. The number of i nt er val s, NI, depends on t he t ot al number of dat a n.
Fol l owi ng equat i on can be used as gui dance f or det er mi ni ng t he sui t abl e number of i nt er val s:
¸ ]
) ( log 3 . 3 1
10
n NI × + · (12.13)
¸ ]
NI denotes that the value is rounded down to the nearest integer.
The l engt h of each i nt er val , LI, i s cal cul at ed usi ng:
, )
¸ ]
NI
x x
LI
min max
−
· (12.14)
wher e x
max
i s t he maxi mum r ecor ded f ai l ur e t i me and x
mi n
i s t he mi ni mu m r ecor d ed f ai l ur e t i me. Th e
l ow er and upper bound of each i nt er val i s cal cul at ed as f ol l ow s:
LI i x X
LI i x X
i
i
× + ·
× − + ·
min max,
min min,
) 1 (
X
min,i
i s t he l ow er bound of t h e i t h i nt er val and X
max,i
i s t he upper bound val ue of t he i t h i nt er val . Let n
1
,
n
2
, … n
n
be t he number of i t ems t hat f ai l i n t he i nt er val i . Then t he est i mat e f or cumul at i ve f ai l ur e
di st r i but i on i s gi ven by
4 . 0
3 . 0
) (
1
max,
^
+
−
·
∑
·
n
n
X F
i
k
k
i
(12.15)
Est i mat e f or t he r el i abi l i t y f unct i on R(t ) i s gi ven by:
4 . 0
7 . 0
) ( 1 ) (
1
max,
^
max,
^
+
+
· − ·
∑
+ ·
n
n
X F X R
n
i k
i
i i
(12.16)
Est i mat e f or t he f ai l ur e densi t y i s gi ven by:
For X
max,i+1
< t < X
max,i
) ( ) 4 . 0 (
) ( ) (
) (
max, 1 max,
1
max, 1 max,
max, 1 max,
^
^
i i
i
i i
i i
X X n
n
X X
X F X F
t f
− × +
·
−
−
·
+
+
+
+
The M TTF i s est i mat ed usi ng t he expr essi on:
221
∑
·
×
·
NI
i
i i med
n
n X
MTTF
1
,
^
(12.17)
wher e X
med,i
i s t he mi dpoi nt i n t he i t h i nt er val and n
k
i s t he number of obser ved f ai l ur es i n t hat i nt er val .
Est i mat e f or sampl e var i ance i s gi ven by
∑
·
× − ·
NI
i
i
i med
n
n
MTTF X s
1
2
^
,
2
) ( (12.18)
Exampl e 12.3
Resul t s of 55 obser ved val ues of t he dur at i on of suppor t t asks i n hour s ar e gi ven i n Tabl e 12.4. Cal cul at e
t he M ean Ti me t o Suppor t (M TTS).
Tabl e 12.4. Ti me t o suppor t dat a
3 56 9 24 56 66 67 87 89 99 4
26 76 79 89 45 45 78 88 89 90 92
99 2 3 37 39 39 77 93 21 24 29
32 44 46 5 46 46 99 47 77 79 89
31 78 34 67 86 86 75 33 55 22 44
SOLUTION:
Fi r st w e need t o f i nd t he number of gr oups usi ng equat i on (12.13). The number of i nt er val s i s gi ven by:
¸ ] ¸ ]
6 74 . 6 ) 55 ( log 3 . 3 1
10
· · × + · NI
The l engt h (r ange) i f each i nt er val (gr oup) i s gi ven by:
¸ ]
17 . 16
6
2 99
min max
·
−
·
−
·
NI
x x
LI
Tabl e 12.5 show s t he var i ous cal cul at i ons associ at ed i n comput i ng t he mean t i me t o suppor t .
Table 12.5. Analysis of grouped data given in example 12.3
i LI (x
min,I
 x
max,i
) n
i
x
med,i
X
med,i
× n
i
1 2  18.17 6 10.08 60.51
2 18.17 34.34 10 26.25 262.55
3 34.34  50.51 11 42.42 466.67
4 50.51  66.68 5 58.59 292.97
5 66.68  82.85 9 74.76 672.88
6 82.85  99 14 90.92 1272.95
M TTS i s gi ven by:
222
∑ ∑
· ·
·
×
·
×
·
6
1
,
1
,
^
06 . 55
55
i
i i med
NI
i
i i med
n X
n
n X
MTTS
12.50. ANALYSIS OF CENSORED DATA
In many cases, t he compl et e dat a may not b e avai l abl e due t o t he r easons such as al l t he i t ems may n ot
have f ai l ed or t he manuf act ur er may w i sh t o get i nt er i m est i mat es of t he r el i abi l i t y et c. The mechani sm
f or censor i ng may be based on a f i xed age, on a f i xed number of f ai l ur es or at some ar bi t r ar y poi nt i n
t i me. In pr act i ce, pr ovi ded t he t i mes at t he t i me of f ai l ur e or , at t he t i me of suspensi on (censor ) ar e
know n, t he r eason f or t er mi nat i ng t he t est i s not i mpor t ant . We w i l l assume t hat t he t i mes of f ai l ur e
ar e know n pr eci sel y. We w i l l l ook at cases i n w hi ch w e do not know t he exact t i me, onl y t hat t he f ai l ur e
occur r ed somet i me bet w een t he l ast i nspect i on and t he cur r ent age l at er . In t hi s sect i on w e der i ve
est i mat es f or f ai l ur e f unct i on, r el i abi l i t y f unct i on when t he dat a i s mul t i pl e censor ed. We denot e t
i
t o
r epr esent a compl et e dat a and t
i
* t o denot e a censor ed t i me.
The onl y di f f er ence bet ween t he est i mat i on of par amet er s i n compl et e dat a and t he censor ed dat a i s
t he cal cul at i on of medi an r anks. Now w e wi l l need t o adj ust t he r anks i n or der t o t ake account of t he
component s t hat have not f ai l ed. The r ank adj ust ment i s done i n t he f ol l owi ng t w o st eps:
1. Sor t al l t he t i mes (f ai l ur es and suspensi ons) i n ascendi ng or der and al l ocat e a sequence number i
st ar t i ng w i t h 1 f or t he f i r st (l ow est ) t i me and endi ng w i t h n (t he sampl e si ze f or t he hi ghest r ecor ded
t i me). Now w e di scar d t he suspended t i mes as i t i s onl y t he (adj ust ed r ank) of t he f ai l ur es w i t h
whi ch we ar e concer ned.
2. For each f ai l ur e cal cul at e t he adj ust ed r ank as f oll ow s:
i
i
i i
S n
R n
R R
− +
− +
+ ·
−
−
2
1
1
1
(12.19)
wher e, R
i
i s t he adj ust ed r ank of t he i
t h
f ai l ur e, R
i1
i s t he adj ust ed r ank of t he (i 1)
t h
f ai l ur e, t hat i s t he
pr evi ous f ai l ur e. R
0
i s zer o and S
i
i s t he sequence number of t he i
t h
f ai l ur e.
As a qui ck check, t he adj ust ed r ank of t h e i
t h
f ai l ur e w i l l al w ays be l ess t han or equal t o t he sequence
number and at l east 1 gr eat er t han t he pr evi ous adj ust ed r ank. If t her e i s no suspensi ons, t he adj ust ed
r ank w i l l be equal t o t he sequence number as bef or e. These adj ust ed r anks ar e t hen subst i t ut ed i nt o
t he Benar d' s appr oxi mat i on f or mul a t o gi ve t he medi an r ank and t he est i mat e f or cumul at i ve pr obabi l i t y
i s gi ven by:
4 . 0
3 . 0
) (
^
+
−
·
n
R
t F
i
i
Exampl e 12.4
The f ol l ow i ng dat a wer e obser ved dur i ng t he dat a capt ur i ng exer ci se on 12 compr essor s t hat ar e bei ng
used by di f f er ent oper at or s. Est i mat e t he r el i abi l i t y and f ai l ur e f unct i on (* i ndi cat es t hat t he dat a i s a
censor ed dat a)
2041, 2173, 2248
*
, 2271, 2567
*
, 2665
*
, 3008, 3091, 3404
*
, 3424, 3490
*
, 3716
223
SOLUTION:
We need t o cal cul at e t he adj ust ed r ank of t he f ai l ur e t i mes usi ng equat i on (12.19), once t hi s i s done,
t hen t he f ai l ur e and r el i abi l i t y f unct i on can be est i mat ed usi ng equat i ons (12.2) and (12.3) r espect i vel y.
The est i mat ed f ai l ur e and r el i abi li t y f unct i ons ar e show n i n Tabl e 12.6.
Tabl e 12.6 Est i mat ed f ai l ur e and r eli abi l i t y f unct i on
S
i
t
i
j R
j
= R
j 1
+ [ (n+1– R
j1
)
/ (n+2 – S
i
) ]
) (
i
t F ) (
i
t R
1 2041 1 1 0.0565 0.9435
2 2173 2 2 0.1370 0.8630
3 2248
*
4 2271 3 3.1 0.2258 0.7742
5 2567
*
6 2665
*
7 3008 4 4.51 0.3395 0.6605
8 3091 5 5.92 0.4532 0.5468
9 3404
*
10 3424 6 7.69 0.5960 0.4040
11 3490
*
12 3716 7 10.34 0.8097 0.1903
12.51. FITTING PROBABILITY DISTRIBUTIONS GRAPHICALLY
The t r adi t i onal appr oach f or measur i ng r el i abi l i t y, mai nt enance and suppor t abi l i t y char act er i st i cs i s
usi ng a t heor et i cal pr obabi l i t y di st r i but i on. It shoul d how ever , be bor ne i n mi nd t hat f ai l ur es do not
occur i n accor dance w i t h a gi ven di st r i but i on. These ar e mer el y conveni ent t ool s t hat can al l ow us t o
make i nf er ences and compar i sons i n not j ust an easi er w ay but al so w i t h know n l evel s of conf i dence. In
t hi s sect i on w e w i l l l ook at a gr aphi cal met hod t hat can be used t o not onl y t o f i t di st r i but i ons t o gi ven
dat a but al so hel p us det er mi ne how good t he f i t i s. To i l l ust r at e t he gr aphi cal appr oach w e use t he
f ol l ow i ng f ai l ur e dat a obser ved on 50 t yr es.
Tabl e 12.7. Fai l ur e dat a f or 50 t yr es
1022 14363 20208 26530 31507
1617 15456 20516 28060 33326
2513 16736 20978 28240 33457
3265 16936 21497 28757 35356
8445 18012 24199 28852 35747
9007 19030 24582 29092 36250
10505 19365 25512 29236 36359
11490 19596 25743 29333 36743
13086 19822 26102 30620 36959
14162 20079 26163 30924 38958
224
To dr aw a gr aph w e obvi ousl y need a set ' x' and ' y' coor di nat es. Sor t i ng t he t i mest of ai l ur e i n
ascendi ng or der w i l l gi ve us t he ' x' val ues so al l w e need i s t o associ at e a cumul at i ve pr obabi l i t y t o each
val ue. Thi s i s done usi ng t he medi an r ank appr oach di scussed ear l i er , t hat i s ' y' axi s val ues ar e gi ven by
t he cumul at i ve f ai l ur e pr obabi l i t i es cal cul at ed usi ng t he equat i on (12.2). Now, w e can pl ot t he val u es [ t
i
,
F(t
i
)] . I n Fi gur e 12.4 w e can see t he r esul t of t hi s f or t he 50 t yr e t i met of ai l ur e.
Fi gur e 12.4 Tyr e Dat a compar ed t o Exponent i al and Nor mal Di st r i but i ons
The t w o addi t i onal l i nes on t hi s gr aph have b een pl ot t ed t o show w hat an exponent i al di st r i but i on (w i t h
t he same mean as t he sampl e) w oul d l ook l i ke and si mi l ar l y f or a nor mal di st r i but i on w i t h t he sampl e
mean and st andar d devi at i on. Thi s i ndi cat es t hat t he exponent i al di st r i but i on i s not a ver y good f i t
w her eas t he nor mal i s cer t ai nl y bet t er . What i t does not t el l us, how ever , i s how much bet t er or ,
i ndeed, w het her anot her di st r i but i on gi ves an even bet t er f i t .
A measur e of how good t he cur ve f i t s t he dat a w oul d be t he cor r el at i on coef f i ci ent but , t hi s onl y appl i es
t o st r ai ght l i ne f i t s. Si mi l ar l y w e coul d use t he Kol mogor ovSmi r nov t est but t hi s r eal l y onl y t el l s us
w het her t he t her e i s a si gni f i cant di f f er ence bet w een t he dat a and t hat w hi ch w oul d b e expect ed i f t he
dat a wer e exponent i al l y or nor mal l y di st r i but ed.
Ther e ar e, i n f act , t w o st andar d appr oaches t o f i t t he dat a t o a pr obabi l i t y di st r i but i on gr aphi cal l y: t o
use “ pr obabi l i t y paper ” or t o t r ansf or m ei t h er t he “ x” or “ y” (or bot h) dat a so t hat t he r esul t i ng gr aph
w oul d be a st r ai ght l i ne i f t he dat a w er e f r om t he gi ven di st r i but i on. Act ual l y bot h met hods ar e
essent i al l y t he same because t o cr eat e pr obabi l i t y paper t he axes have been so const r uct ed as t o
pr oduce st r ai ght l i nes pl ot i f t he dat a i s f r om t he gi ven di st r i but i on. If w e can det er mi ne t he necessar y
t r ansf or ms t hen w e can easi l y const r uct t he pr obabi l i t y paper .
Fitting an exponential distribution to data graphically
The cumul at i ve pr obabi l i t y densi t y f unct i on f or t he exponent i al di st r i but i on i s gi ven by
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 5000 10000 15000 20000 25000 30000 35000 40000
TimetoFailure
P
r
o
b
a
b
i
l
i
t
y
Tyre Data
Exponential
Normal
225
¹
'
¹
≥ − −
<
·
0 ), exp( 1
0 , 0
) (
t t
t
t F
λ
Si nce we ar e onl y consi d er i ng posi t i ve f ai l ur e t i mes, we can, wi t hout l oss of gener al i t y, omi t t he
expr essi on f or t < 0. If we r epl ace F(t ) w i t h p t hen we get
) exp( 1 t p λ − − ·
Rear r angi ng and t aki ng nat ur al l ogar i t hm we get
t
p
λ ·
−
]
1
1
ln[ (12.20)
Thi s i s a l i near f unct i on i n t such t hat t he sl ope of t he l i ne i s t he r eci pr ocal of t he M TTF. Fi gur e 12.5 i s an
exampl e of “ Exponent i al Gr aph Paper ” (f or t he f ai l ur e dat e f r om Tabl e 12.7). The yscal e i s gi ven as
per cent ages r at her t han pr obabi l i t i es. The xscal e i s l i near .
Fi gur e 12.5 Dat a Pl ot t ed on Exponent i al Gr aph Paper
If t he dat a f or ms a st r ai ght l i ne i n t he exponent i al pr obabi l i t y paper , t hen w e can f i nd t he val ue of M TTF
by usi ng t he r el at i on F(M TTF) = 0.632. That i s, w e f i nd t he t i me t o f ai l ur e f r om t he paper f or w hi ch t h e
per cent age f ai l ur es i s 63.2.
0 5000 10000 15000 20000 25000 30000 35000 40000
Times to Failure
C
u
m
u
l
a
t
i
v
e
P
r
o
b
a
b
i
l
i
t
y
(
%
)
99
95
90
80
70
60
50
40
20
226
Fitting a Normal Distribution Graphically
We w i l l now see how good a f i t t he nor mal di st r i but i on gi ves. Agai n w e can pl ot t he t i mest of ai l ur e on
speci al nor mal (pr obabi l i t y) paper . Such paper i s becomi ng i ncr easi ngl y mor e di f f i cul t t o obt ai n
commer ci al l y. It can, however , be cr eat ed usi ng a pr o pr i et ar y spr eadsheet package. Fi gur e 12.6 shows
how t he t yr e exampl e f ai l ur e t i mes (and t h ei r r espect i ve medi an r anks) woul d appear on “ nor mal
paper ” .
Fi gur e 12.6 Ti mes t oFai l ur e pl ot t ed on Nor mal Paper
The cumul at i ve densi t y f unct i on f or t he nor mal di st r i but i on i s not as si mpl e t o t r ansf or m t o a l i near f or m
as t he exponent i al .
dx e p t F
t
x
∫
∞ −
,
_
¸
¸ −
−
· ·
2
2
1
2
1
) (
σ
µ
σ π
How ever , w e can obt ai n t he st andar di sed nor mal var i abl e ) (
σ
µ −
·
x
z , f or any gi ven val ue of p (F(t ))
ei t her f r om t abl es or , usi ng t he NORM SINV f unct i on i n M i cr oSof t ™ Excel ®, f or exampl e. Now w e can
pl ot t hi s val ue(as t he y coor di nat e) agai nst t he cor r espondi ng t i met of ai l ur e (as t he x coor di nat e). Th e
val ue of µ and σ can be f ound by usi ng t he r el at i on, F(µ) = 0.5 and F(µ+σ) = 0.84.
Fitting a LogNormal Distribution Graphically
Essent i al l y t he l ognor mal di st r i but i on i s t he same as a nor mal di st r i but i on except i ng t hat t he (nat ur al )
l ogar i t hm of t h e x val ues ar e used i n pl ace of t he act ual val ues. Fi gur es 12.7 and 12.8 show l ognor mal
pl ot f or t he dat a gi ven i n Tabl e 12.7.
Times to Failure
0
10000 20000 30000 40000
99
95
90
80
70
60
50
40
30
20
10
5
1
227
Fi gur e 12.7 Ti mest oFai l ur e pl ot t ed on LogNor mal Paper
Fi gur e 12.8 Fi t t i ng a LogNor mal Di st r i but i on Gr aphi cal l y
1000 10000 100000
99
95
90
80
70
60
50
40
30
20
10
5
1
y = 1.0614x  10.444
R
2
= 0.7346
4
3
2
1
0
1
2
3
6 6.5 7 7.5 8 8.5 9 9.5 10 10.5 11
Ln(Time to Failure)
N
o
r
m
a
l
O
r
d
i
n
a
t
e
228
Her e t he pl ot t ed poi nt s f or m a concave cur ve t o w h i ch t he st r ai ght l i ne i s not a par t i cul ar l y good f i t
al t hough i t i s st i l l bet t er t han t he exponent i al f i t . The mean i n t hi s case i s 18,776 w hi ch i s consi der abl y
l ow er t han t he mean f r om t he pr evi ous gr aphs but , t hi s i s because i t i s t he geomet r i c mean (t he n
t h
r oot
of t he pr oduct of t he TTFs) and not t he ar i t hmet i c mean w i t h w hi ch w e ar e mor e f ami l i ar .
Fitting a Weibull Distribution Graphically
The cumul at i ve densi t y f unct i on of t he Wei bul l di st r i but i on i s si mi l ar t o t hat of t he exponent i al , i ndeed
t he l at t er i s t he (mat hemat i cal l y) degener at i ve f or m of t he f or mer .
By r ear r angi ng and t aki ng nat ur al l ogar i t hms
whi ch i s st il l not i n a l i near f or m so we have t o t ake l ogs agai n t o gi ve:
So i f w e pl ot l n(l n(1p)) agai nst l n(t ) an est i mat e of t he shape par amet er (β) of t he Wei bul l w i l l be gi ven
by t he sl ope of t he st r ai ght l i ne dr aw n t hr ough t he pl ot t ed poi nt s. To get an est i mat e of t he scal e
par amet er (η) w e need t o car r yout a t r ansf or m on t he i nt er cept :
wher e c i s t he i nt er cept of t he r egr essi on l i ne w i t h t he xaxi s. Fi gur es 12.9 and 12.10 show s Wei bul l pl ot
f or t he dat a gi ven i n Tabl e 12.7.
F t p ( ) · ·
≥
¹
'
¹
¹
¹
¸
¸
_
,
0 for t < 0
1 for t 0 e

t
η
β
− − ·
¸
¸
_
,
ln( ) 1 p
t
η
β
ln( ln( )) ln( ) ln( ) − − · − 1 p t β β η
η
β
·
−
e
c
229
Figure 12.9 TimestoFailure Fitted on Weibull Paper
Fi gur e 12.10 Fi t t i ng a Wei bul l di st r i but i on gr aphi cal l y
Agai n t he Wei bul l di st r i but i on does not gi ve as good a f i t as t he nor mal (di st r i but i on) but i t i s bet t er t han
ei t her t he exponent i al or t he l ognor mal . The sl ope (1.48) i ndi cat es t hat t her e coul d be a cer t ai n
amount of ager el at edness t o t he f ai l ur es.
99
95
90
80
70
6
50
40
30
20
10
5
1
1000 10000 100000
y = 1.4608x  14.935
R
2
= 0.883
5
4
3
2
1
0
1
2
6 6.5 7 7.5 8 8.5 9 9.5 10 10.5 11
Log(Time to Failure)
T
r
a
n
s
f
o
r
m
e
d
P
r
o
b
a
b
i
l
i
t
y
230
12.52. REGRESSION
The model s used t o r el at e a dependent var i abl e y t o t he i ndependent var i abl es x ar e cal l ed r egr essi on
model s. The si mpl est r egr essi on model i s t he one t hat r el at es t he var i abl e y t o a si ngl e i ndependent
var i abl e x (l i near r egr essi on model ). Li near r egr essi on pr ovi des pr edi ct ed val u es f or t he dependent
var i abl es (y) as a l i near f unct i on of i ndependent var i abl e (x). That i s, l i near r egr essi on f i nds t he best f i t
st r ai ght l i ne f or t he set of poi nt s (x, y). The obj ect i ves of l i near r egr essi on ar e:
1. To check whether there is a linear relationship between the dependent variable and the independent
variable.
2. To find the best fit straight line for a given set of data points.
3. To estimate the constants a` and b` of the best fit y = a + bx.
Fi gur e 12.11 Least squar e r egr essi on.
The st andar d met hod f or l i near r egr essi on anal ysi s (f i t t i ng a st r ai ght l i ne t o a si ngl e i ndependent
var i abl e) i s usi ng t he met hod of l east squar es. Least squar e r egr essi on i s a pr ocedur e f or est i mat i ng t he
coef f i ci ent s ‘ a’ and ‘ b’ f r om a set of X, Y poi nt s t hat have been measur ed. In r el i abi l i t y anal ysi s, t he set X
i s t he set of t i me t o f ai l ur es (or f unct i on of TTF) and set Y i s t hei r cor r espondi ng cumul at i ve pr obabi l i t y
val ues (or f unct i on of cumul at i ve di st r i but i on). Fi gur e 12.11 i l l ust r at es t he l east squar e r egr essi on. The
measur e of how w el l t hi s l i ne f i t s t he dat a i s gi ven by t he cor r el at i on coef f i ci ent . If w e const r uct a l i ne
such t hat i t passes t hr ough t he poi nt , ) y x, wher e x i s t he mean of t he x val ues and y i s t he mean of
t he y val ues t hen t he sum of t he di st ances bet w een each poi nt and t he poi nt on t he l i ne ver t i cal l y above
(ve) or bel ow (+ve) wi l l al ways be zer o (pr o vi ded t he l i ne i s not par al l el t o t he yaxi s). The same hol ds
f or t he hor i zont al di st ances pr ovi ded t hat t he l i ne i s not par al l el t o t he xaxi s. Thi s means t hat any l i ne
passi ng t hr ough t he means (i n t he way descr i bed) wi l l be an unbi ased est i mat or of t he t r ue l i ne.
If w e now assume t hat t her e i s a l i near r el at i onshi p bet w een t he x’ s (x ∈ X) and y’ s ( y ∈ Y), t hat t he x’ s
ar e know n exact l y and t hat t he “ er r or s” i n t he y val u es ar e nor mal l y di st r i but ed w i t h mean 0 t hen i t can
be shown t hat t he val ues of a and b w hi ch mi ni mi ses t he expr essi on:
∑
·
− −
n
i
i i
bx a y
1
2
) ( (12.21)
y
a
x
Slope = b
231
Wi l l gi ve t he best f i t . The expr essi on (y
i
– a – bx
i
) gi ves t he ver t i cal di st ance bet w een t he poi nt and t he
l i ne. Cut t i ng out l ot of al gebr a, one can show t hat t he val ues of a and b can be f ound by sol vi ng t he
f ol l ow i ng equat i ons:
∑ ∑
· ·
· +
n
i
i
n
i
i
y x b na
1 1
(12.21)
∑ ∑ ∑
· · ·
· +
n
i
i i
n
i
i
n
i
i
y x x b x a
1 1
2
1
(12.22)
‘ a’ i s t he est i mat e of t he i nt er cept (of t he l i ne w i t h t he yaxi s) and ‘ b’ i s t he est i mat e of t he sl ope – i .e. y
= a + bx i s t he equat i on of t he l i ne gi vi ng:
2
1 1
2
1 1 1
) (
∑ ∑
∑ ∑ ∑
· ·
· · ·
−
−
·
n
i
i
n
i
i
n
i
i
n
i
i
n
i
i i
x x n
y x y x n
b (12.23)
∑ ∑
· ·
− ·
n
i
i
n
i
i
n
x
b
n
y
a
1 1
(12.24)
Not e al so t hat t hese expr essi ons ar e not symmet r i cal i n x and y. The f or mul a quot ed her e gi ves
w hat i s cal l ed “ y on x” r egr essi on and i t assumes t he er r or s ar e i n t he yval ues.
By r epl aci ng each x w i t h a y and each y w i t h an x w e can per f or m “ x on y” r egr essi on (w hi ch assumes
t he er r or s ar e i n t he x val ues). If c i s t he est i mat e of t he i nt er cept so obt ai ned and d i s t he est i mat e
of t he sl ope t hen t o get est i mat es of a and b (t he i nt er cept and sl ope of t he or i gi nal gr aph):
d
b
1
· and
d
c
a − ·
Not e: unl ess t he poi nt s ar e col l i near , t he “ x on y” est i mat es w i l l not be t he same as t he “ y on x”
est i mat es. In t he sp eci al case wh er e you want t o f or ce t he l i ne t hr ough t he or i gi n (i .e. t he i nt er cept i s
zer o), t he l east squar es f or mul a f or t he sl ope becomes:
∑
∑
·
·
·
n
i
i
n
i
i i
x
y x
b
1
2
1
(12.25)
Not e t hi s l i ne does not pass t hr ough t he means (unl ess i t i s a per f ect f i t ).
232
Correlation Coefficient
A measur e of t he dependence bet ween t wo var i abl es i s gi ven by t he cor r el at i on coef f i ci ent . The
cor r el at i on coef f i ci ent , r i s gi ven by:
2
1 1
2 2
1 1
2
1 1 1
) ( ) ( ∑ ∑ ∑ ∑
∑ ∑ ∑
· · · ·
· · ·
− × −
−
·
n
i
i
n
i
i
n
i
i
n
i
i
n
i
i
n
i
i
n
i
i i
y y n x x n
y x y x n
r (12.26)
The cor r el at i on coef f i ci ent al w ays l i es bet w een –1 and +1. A val ue of +1 or –1 means t hat x and y ar e
exact l y l i near l y r el at ed. In t he f or mer case y i ncr eases as x i ncr eases but f or r = 1, y decr eases as x
i ncr eases. Not e t hat i f x and y ar e i ndependent t hen r = 0, but r = 0 does not mean t hat x and y ar e
i ndependent . The best f i t di st r i but i on i s t he one w i t h maxi mum r val ue (cl ose t o one). To f i nd t he best
f i t , r egr essi on anal ysi s i s car r i ed out on t he popul ar di st r i but i on such as exponent i al , Wei bul l , nor mal
and l ognor mal . The one w i t h hi ghest cor r el at i on coef f i ci ent i s sel ect ed as t he best . The coor di nat es (x,
y) and t he cor r espondi ng par amet er s f or di f f er ent di st r i but i ons ar e l i st ed gi ven i n t he f ol l ow i ng sect i ons.
Linear Regression for Exponential Distribution
To f i t a dat a t o an exponent i al di st r i but i on, w e t r ansf or m t he coor di nat es (t
i
, F(t
i
)) such a w ay t hat ,
when pl ot t ed, i t gi ves a st r ai ght l i ne. Her e t
i
i s t he obser ved f ai l ur e t i mes and F(t
i
) i s t he est i mat ed
cumul at i ve di st r i but i on f unct i on. The cumul at i ve di st r i but i on of exponent i al di st r i but i on i s gi ven by:
) exp( 1 ) ( t t F λ − − ·
t hat i s,
t
t F
λ ·
−
]
) ( 1
1
ln[ (12.27)
Equation (12.27) is a linear function. Thus, for an exponential distribution, the plot of ( ]
) ( 1
1
ln[ ,
t F
t
−
)
provides a straight line. Thus, if t
1
, t
2
, ., t
n
are the observed failure times, then to fit this data into an
exponential distribution, we set:
i i
t x · (12.28)
]
) ( 1
1
ln[
i
i
t F
y
−
· (12.29)
Subst i t ut i ng (x
i
, y
i
) i n equat i on (12.23) we get :
233
∑
∑
·
·
·
n
i
i
n
i
i i
x
y x
b
1
2
1
(12.30)
Not e t hat , f or exponent i al di st r i but i on b = 1/ M TTF.
Exampl e 12.5
The following failure data were observed on Actuators. Fit the data to an exponential distribution and find
the MTTF and the correlation coefficient.
14, 27, 32, 34, 54, 57, 61, 66, 67, 102, 134, 152, 209, 230
SOLUTION:
First we carry out least square regression on ]
) ( 1
1
ln[ ,
i
i
t F
t
−
, various calculations are tabulated in Table
12.8.
Tabl e 12.8. Regr essi on anal ysi s f or t he dat a i n exampl e 12.5
i t
i
(= x
i
) F(t
i
) y
i
= l n[ 1 / (1F(t
i
))]
1 14 0.0486 0.0498
2 27 0.1180 0.1256
3 32 0.1875 0.2076
4 34 0.2569 0.2969
5 54 0.3263 0.3951
6 57 0.3958 0.5039
7 61 0.4652 0.6260
8 66 0.5347 0.7651
9 67 0.6041 0.9267
10 102 0.6736 1.1196
11 134 0.7430 1.3588
12 152 0.8125 1.6739
13 209 0.8819 2.1366
14 230 0.9513 3.0239
The val ue of b i s gi ven by:
01126 . 0
]
) ( 1
1
ln[
1
2
1
1
2
1
·
−
×
· ·
∑
∑
∑
∑
·
·
·
·
n
i
i
n
i i
i
n
i
i
n
i
i i
t
t F
t
x
y x
b
MTTF is given by 1/b = 1/0.01126 = 88.73. The corresponding correlation coefficient is 0.9666.
234
Linear Regression for Weibull Distribution
Cumul at i ve di st r i but i on of Wei bul l di st r i but i on i s gi ven by:
) ) ( exp( 1 ) (
β
η
t
t F − − ·
That is, ) ln( ) ln( )]
) ( 1
1
ln[ln( η β β − ·
−
t
t F
, which is a linear function. Thus to fit the data to a Weibull
distribution, we set:
) ln(
i i
t x · (12.31)
)]
) ( 1
1
ln[ln(
i
i
t F
y
−
· (12.32)
From least square regression, it is evident that the shape and scale parameters of the distribution are given
by:
b · β (12.34)
) / exp( β η a − · (12.35)
Exampl e 12.6
Const r uct a l east squar e r egr essi on f or t he f ol l owi ng f ai l ur e dat a:
17, 21, 33, 37, 39, 42, 56, 98, 129, 132, 140
SOLUTION:
M aki ng use of equat i ons (12.31) and (12.32), w e const r uct t he l east squar e r egr essi on, w hi ch ar e
pr esent ed i n Tabl e 12.9.
Tabl e 12.9. Wei bul l r egr essi on f or t he dat a i n exampl e 12.6
i t
i
F(t
i
) x
i
= l n(t
i
) Y
i
= l nl n(1/ 1F(t
i
))
1 17 0.0614 2.8332  2.7581
2 21 0.1491 3.0445  1.8233
3 33 0.2368 3.4965  1.3082
4 37 0.3245 3.6109  0.9354
5 39 0.4122 3.6635  0.6320
6 42 0.5 3.7376  0.3665
7 56 0.5877 4.0253  0.1209
8 98 0.6754 4.5849 0.1180
9 129 0.7631 4.8598 0.3648
10 132 0.8508 4.8828 0.6434
11 140 0.9385 4.9416 1.0261
235
Using equations (12.34) and (12.35), we get β = 1.4355, η = 76.54 and the correlation
coefficient r = 0.9133.
Linear regression for Normal Distribution
For nor mal d i st r i but i on,
) ( ) ( ) ( z
t
t F Φ ·
−
Φ ·
σ
µ
Now z can be written as:
σ
µ
σ σ
µ
− ·
−
· Φ ·
− i i
i
t t
t F z )] ( [
1
(12.36)
Which is a linear function. Now for regression, we set x
i
= t
i
and y
i
= z
i
= Φ
1
[F(t
i
)]. The value of z can
be obtained from standard normal distribution table. One can also use the following expression that gives
polynomial approximation for z
i
.
i i
t x · (12.37)
]
)] ( 1 [
1
ln[
2
i
t F
P
−
·
3
3
2
2 1
2
2 1 0
1 P d P d P d
P C P C C
P y
i
+ + +
+ +
− · (12.38)
where
C
0
= 2.515517, C
1
= 0.802853, C
2
= 0.010328, d
1
= 1.432788,
d
2
= 0.189269, d
3
= 0.001308
The est i mat e f or µ and σ ar e gi ven by
b
a
− · µ and
b
1
· σ
Exampl e 12.7
Fi t t he f ol l owi ng dat a i nt o a nor mal di st r i but i on
236
62, 75, 93, 112, 137, 170, 185
SOLUTION:
Tabl e 12.10 gi ves var i ous comput at i ons i nvol ved i n r egr essi on.
Tabl e 12.10. Nor mal r egr essi on f or exampl e 12.7
i t
i
F(t
i
) z
i
= P – (c
0
+ c
1
P + c
2
P
2
/ 1 + d
1
P + d
2
P
2
+
d
3
P
3
)
1 62 0.0945  1.2693
2 75 0.2297  0.7302
3 93 0.3648  0.3434
4 112 0.5 0
5 137 0.6351 0.3450
6 170 0.7702 0.7394
7 185 0.9054 1.3132
The est i mat e f or µ = 118.71, σ = 54.05 and t he cor r el at i on coef f i ci ent r = 0.9701.
Linear Regression for Lognormal Distribution
For l ognor mal di st r i but i on w e set :
) ln(
i i
t x · (12.39)
]
)] ( 1 [
1
ln[
2
i
t F
P
−
·
3
3
2
2 1
2
2 1 0
1 P d P d P d
P C P C C
P y
i
+ + +
+ +
− · (12.40)
where
C
0
= 2.515517, C
1
= 0.802853, C
2
= 0.010328, d
1
= 1.432788, d
2
= 0.189269, d
3
= 0.001308
237
Total Cost of Ownership (TCO) and Life Cycle Cost
Models
Prof. U Dinesh Kumar
Indian Institute of Management Bangalore
238
CONTENTS
1. Life Cycle Cost and Total Cost of Ownership 3
1.1 Introduction. 3
1.2 Initial Capital Cost 5
1.3 Life of the Asset 5
1.4 The Discount Rate 6
1.5 Operation and Maintenance Cost 6
1.6 Disposal Cost 6
1.7 Uncertainties and Sensitivity Analysis 6
1.8 Summary 7
2. Survey of Existing Literature 09
2.1 Life Cycle Cost/Total Cost of Ownership Concept 09
2.2 Definitions of Life Cycle Cost/Cost of Ownership 12
2.3 Life Cycle Cost Technique 13
2.4 Review of Life Cycle Cost Models 16
2.5 Taylor`s LCC Model 23
2.6 Raymer`s LCC Model 25
2.7 Roskam`s LCC Model 25
2.8 Fabrycky and Blanchard`s LCC Model 26
2.9 Burn`s Life Cycle Cost Model 28
2.10 PRICE Life Cycle Costing System 28
2.11 Equipment Designer`s Cost Analysis System (EDCAS) 28
2.12 LCC Models Using Markov Chain 30
2.13 LCC Models for Labour Factor 31
2.14 LCC Models in Designing for Logistic Support 31
2.15 Applications of LCC/Cost of Ownership Models 34
References 35
239
240
1. Life Cycle Cost and Total Cost of Ownership
___________________________________________
1.1 Int r oduct i on
“ Val ue f or M oney” has become one of t he i mpor t ant cr i t er i a i n an i ncr easi ngl y compet i t i ve
busi ness envi r onment . Li f e Cycl e Cost (LCC) and t he Tot al Cost of Owner shi p (TCO) ar e t wo i mpor t ant
f i nanci al measur es t hat ar e used f or deci si on maki ng i n acqui si t i ons. Fr om i t s or i gi ns i n def ence
equi pment pr ocur ement i n t he US i n ear l y 1960s, t he use of l i f e cycl e cost and cost of ow ner shi p has
ext ended t o ot her ar eas of t he publ i c and pr i vat e sect or s. LCC and TCO ar e bei ng used t o assi st i n
deci si onmaki ng, budget pl anni ng, cost cont r ol , and r ange of ot her act i vi t i es t hat occur over t he l i f e of
compl ex t echnol ogi cal equi pment .
It i s i mpor t ant t o consi der t he di f f er ence bet w een LCC and TCO. LCC anal ysi s i s appl i ed
r out i nel y t o mi l i t ar y pr oj ect s. In t he mi l i t ar y sect or t he consumer , by f undi ng t he pr oj ect and oper at i n g
t he r el at ed pr oduct , essent i al l y bear s t he t ot al l i f e cycl e cost cover i ng t he maj or cost el ement s i n al l
st ages of a pr oduct ’ s l i f e cycl e. The t er m LCC anal ysi s i s r ar el y used i n t he commer ci al sect or . Inst ead,
t he mai n f ocus i s on TCO wh er e r el at ed cost s, cover i ng acqui si t i on (pur chase or l ease), oper at i on,
mai nt enance and suppor t ar e bor ne by t he cust omer . In addi t i on, t he cust omer can al so i ncur cost s
w hen t he pr oduct i s not avai l abl e f or use, t hat i s, ‘ down t i me cost s’ .
The obj ect i ves of LCC/ TCO ar e (Fl anagan and Nor man, 1983):
Ø To enabl e i nvest ment opt i ons t o be mor e ef f ect i vel y eval uat ed.
Ø To consi der t he i mpact of all cost s r at her t han onl y t he i ni t i al capi t al cost s.
Ø To assi st i n t he ef f ect i ve management of compl et ed pr oj ect s.
241
Ø To f aci l i t at e choi ce bet w een compet i ng al t er nat i ves.
In t he Def ence i ndust r y t he syst em’ s l i f e cycl e i s di vi ded i nt o var i ous phases, w hi ch al l ow pr oper
pl anni ng and cont r ol of a pr oj ect . The number of phases depend on t he nat ur e of t he pr oj ect , pur pose
and whet her t hey ar e appl i ed t o commer ci al , mi l i t ar y or space pr oj ect s (Knot t s, 1998). Commonl y used
phases ar e:
1. Requi r ement s (Funct i onal Speci f i cat i on).
2. Concept / Feasi bi l i t y St udi es.
3. Desi gn and Devel op ment .
4. Pr oduct i on.
5. Test i n g and Cer t i f i cat i on.
6. Oper at i on, M ai nt enance and Suppor t .
7. Di sposal
It i s r epor t ed by t he US Depar t ment of Def ence t hat 70% of w eapon syst em l i f e cycl e cost i s
commi t t ed by t he end of concept st udi es, 85% by t he end of syst em def i ni t i on and 95% by t he end of
f ul l scal e devel opment . The US Depar t ment of Def ence has f or mal l y used t he concept of l i f e cycl e cost
i n w eapon syst em acqui si t i on si nce t he ear l y 1960s t hr ough l i f e cycl e cost i ng and l i f e cycl e cost anal ysi s.
The cost of ow ner shi p appr oach i d ent i f i es al l f ut ur e cost s and r educes t hem t o t hei r pr esent
val ue by use of t he di scoun t i ng t echni ques t hr ough w hi ch t he economi c w or t h of a pr oduct or pr oduct
opt i ons can be assessed. In or der t o achi eve t hese obj ect i ves t he f ol l ow i ng el ement s of cost of
owner shi p have been i dent i f i ed (Woor war d, 1997):
Ø Ini t i al capi t al cost s
Ø Li f e of t he asset
Ø The di scount r at e
242
Ø Oper at i ng and mai nt enance cost s
Ø Di sposal cost
Ø Uncer t ai nt y and sensi t i vi t y anal ysi s
1.2 Ini t i al capi t al cost s
The i ni t i al capi t al cost s can be di vi ded i nt o t hr ee subcat egor i es of cost namel y: (1) pur chase cost s,
(2) acqui si t i on/ f i nance cost s, and (3) i nst al l at i on/ commi ssi oni ng/ t r ai ni ng cost s. Pur chase cost s w i l l
i ncl ude assessment of i t ems such as l and, bui l di ngs, f ees, and equi pment . Fi nance cost s i ncl ude
al t er nat i ve sour ces of f unds. Basi cal l y, t he i ni t i al capi t al cost cat egor y i ncl udes al l t he cost s of buyi ng
t he physi cal asset and br i ngi ng i t i nt o oper at i on.
1.3 Lif e of t he Asset
The est i mat ed l i f e of an asset has a maj or i nf l uence o n l i f e cycl e cost anal ysi s. Fer r y et al (1991) has
def i ned t he f ol l ow i ng f i ve possi bl e det er mi nant s of an asset ’ s l i f e expect ancy:
Funct i onal l i f e – t he per i od over w hi ch t he need f or t he asset i s ant i ci pat ed.
Physi cal l i f e – t he per i od over w hi ch t he asset may be expect ed t o l ast physi cal l y, t o w hen
r epl acement or maj or r ehabi l i t at i on i s physi cal l y r equi r ed.
Technol ogi cal l i f e – t he per i od unt i l t echni cal obsol escence di ct at es r epl acement due t o t he
devel op ment of a t echnol ogi cal l y super i or al t er nat i ve.
Economi c l i f e – t he per i od unt i l economi c obsol escence di ct at es r epl acement w i t h a l ow er cost
al t er nat i ve.
Soci al and l egal l i f e – t he per i od unt i l human desi r e or l egal r equi r ement di ct at es r epl acement .
1.4 The discount r at e
243
As t he cost of ow ner shi p i s di scount ed t o t hei r pr esent val ue, sel ect i on of a sui t abl e di scount r at e i s
cr uci al f or TCO anal ysi s. A hi gh di scount r at e w i l l t end t o f avour opt i ons w i t h l ow capi t al cost , shor t l i f e
and hi gh r ecur r i ng cost , whi l st a l ow di scount r at e wi l l have t he opposi t e ef f ect .
1.5 Oper at i ons and M ai nt enance Cost s
Cost of ow ner shi p, i n many cases, i s about oper at i on and mai nt enance cost . Est i mat i on of
oper at i on and mai nt enance cost s i s t he essent i al t o mi ni mi se t he t ot al cost of ow ner shi p of t he asset . I n
t he whol e of TCO anal ysi s, est i mat i on of oper at i on and mai nt enance i s t he most chal l engi ng t ask.
1.6 Di sposal cost
Thi s i s t he cost i ncur r ed at t he end of as asset ’ s w or ki ng l i f e i n di sposi ng of t he asset . The
di sposal cost woul d i ncl ude t he cost of demol i t i on, scr appi ng or sel l i ng t he asset .
1.7 Uncer t ai nt i es and Sensi t i vi t y Analysi s
LCC/ TCO i s hi ghl y dependent on t he assu mpt i ons and est i mat es made whi st col l ect i ng dat a.
Even t hough i t i s possi bl e t o i mpr o ve t he qual i t y of t hese est i mat es, t her e i s al ways an el ement of
uncer t ai nt y associ at ed wi t h t hese est i mat es and assumpt i ons. M acedo et al (1978) i dent i f i es t h e
f ol l ow i ng f i ve maj or sour ces of uncer t ai nt y:
1. Di f f er ences b et w een t he act ual and expect ed per f or mance of t he syst em coul d af f ect f ut ur e
oper at i on and mai nt enance cost .
2. Changes i n oper at i onal assumpt i ons ari si ng f r om modif i cat i ons i n user act i vi t y.
3. Fut ur e t echnol ogi cal advances t hat coul d pr ovi d e l ow er cost al t er nat i ves and hence shor t en t he
economi c l i f e of any syst em/ subsyst em.
4. Changes i n t he pr i ce l evel s of maj or r esour ces such as ener gy or manpower , r el at i ve t o ot her
r esour ces can af f ect f ut ur e al t er at i on cost s.
244
5. Er r or i n est i mat i ng r el at i onshi ps, pr i ce r at es f or speci f i c r esour ces and t he r at e of i nf l at i on i n
over al l cost s f r om t he t i me of est i mat i on t o t he avai l abi l i t y of t he asset .
Whi l e under t aki ng a LCC/ TCO anal ysi s, t her e may be some key par amet er s about whi ch uncer t ai nt y
exi st s, usual l y because of t he i nadequacy of t he i npu t dat a. Bl anchar d (1972) suggest ed t he f ol l owi ng
shoul d be t he subj ect of sensi t i vi t y anal ysi s:
Ø Fr equency of t he mai nt enance f act or .
Ø Var i at i on of t he asset ’ s ut i l i zat i on or oper at i ng t i me.
Ø Ext ent of t he syst em’ s sel f di agnost i c capabi l i t y.
Ø Var i at i on of cor r ect i ve mai nt enance hour s per oper at i ng hour .
Ø Pr oduct demand r at e.
Ø The di scount r at e
1.8 Summar y
In t hi s chapt er , w e l ooked at t he concept of Li f e Cycl e Cost and Cost of Ow ner shi p and t he
f act or s t hat i nf l uence LCC and TCO as descr i bed i n t he l i t er at ur e. In t he next chapt er w e sur vey t he
exi st i ng LCC/ TCO model s, met hodol ogi es, pr act i ces an d t echni ques avai l abl e i n t he l i t er at ur e and i t s
appl i cat i ons and l i mi t at i ons. The model s t hat have di r ect appl i cat i on f or assessi ng t he t ot al cost of
owner shi p of ai r bor ne mi l i t ar y equi pment ar e hi ghl i ght ed.
245
53. 2. SURVEY OF EXISTING LITERATURE
Al t hough a consi der abl e body of l i t er at ur e r el at i ng t o l i f e cycl e cost and cost of ow ner shi p has
been devel oped over t he past f our decades, much of t he publ i shed mat er i al has emanat ed f r o m
pr act i t i oner s (Ni chol as, 1999). Publ i cat i ons by pr act i t i oner s have t ended t o consi st of gener al gui del i nes
and a subst ant i al amount of t echni cal r epor t s i n t he f or m of t echni cal r epor t s and conf er ence paper s,
det ai l i ng t he devel opment and appl i cat i on of speci f i c model s and model i ng t echni ques. M ost of t hese
paper s l acked r i gor t hat one can expect f r om academi c publ i cat i ons. M u ch f ew er i n number , academi c
publ i cat i ons have t aken t h e f or m of t ext books w hi ch pr esent t ool s and t echni ques of anal ysi s (Dhi l l on
1989, Fabr ycky and Bl anchar d 1991) and Jour nal publ i cat i ons, w hi ch t end t o consi der ver y speci f i c
t echni cal aspect s of LCC and TCO. We have gr ouped t he l i t er at ur e under di f f er ent cl assi f i cat i on, namel y,
(1) Publ i cat i ons on LCC/ TCO concept s, (2) Publ i cat i ons on LCC model s, (3) Publ i cat i ons on TCO model s,
and (4) Publ i cat i ons on LCC/ TCO appl i cat i ons.
2.1 Li f e Cycl e Cost / Tot al Cost of Ownershi p Concept
Asi edu and Gu (1998) i n t hei r paper t i t l ed, ‘ Pr oduct l i f e cycl e cost anal ysi s – st at e of t he ar t
r evi ew ,’ pr ovi des an i nd ept h anal ysi s of sever al i ssues of t he l i f e cycl e cost . The paper di scusses i ssues
such as (1) l i f e cycl e appr oach t o desi gn; (2) l i f e cycl e cost anal ysi s, and (3) cost anal ysi s model s. They
poi nt out t hat LCC anal ysi s shoul d not be seen as an appr oach f or d et er mi ni ng t he cost of t he syst em
but as an ai d t o desi gn deci si onmaki ng. The use of l i f e cycl e cost anal ysi s and cost of ow ner shi p shoul d
246
t her ef or e be r est r i ct ed t o t he cost t hat w e can cont r ol . For desi gner s, est i mat i ng t he LCC of a pr oposed
pr oduct dur i ng i t s devel opment phase i s r equi r ed f or a number of r easons i ncl udi ng:
(1) Det er mi ni ng t he most cost ef f i ci ent desi gn amongst a set of al t er nat i ves.
(2) Det er mi ni ng t he cost of a desi gn f or budget ar y pur poses.
(3) Ident i f yi ng cost dr i ver s f or desi gn changes and opt i mi sat i on.
Fi gur e 1. Key f act or s i n Li f e Cycl e Cost (Rose, 1984)
Rose (1984) i n a shor t paper r evi ew i ng t he st at us of l i f e cycl e cost ar gues t hat many f or ms of l i f e
cycl e cost anal ysi s ar e not act ual l y l i f e cycl e cost anal ysi s but can cont r i but e t o l i f e cycl e cost anal ysi s.
Thi s i s an i mpor t ant di st i nct i on because w i de r anges of anal yses ar e of t en t er med as l i f e cycl e cost
anal ysi s, wher eas t hey ar e par t i al anal ysi s.
Rose concl udes t hat many l i f e cycl e cost st udi es r el at e t o par t of a syst em r at her t han a
compl et e syst em. He has shown t he r el at i onshi p bet ween capi t al and r evenue cost s, t he pot ent i al
Life cycle costing
functions
Physical asset
functions
Financial
functions
Organizational
functions
Specification
Costs
Time
Tradeoff
Capital cost
Revenue Cost
Design
Development
Acquisition
Disposal
Operations
Maintenance
Reliability
Availability
Maintainability
Capacity
Utility
Performance
Life cycle costing
functions
Physical asset
functions
Financial
functions
Organizational
functions
Specification
Costs
Time
Tradeoff
Capital cost
Revenue Cost
Design
Development
Acquisition
Disposal
Operations
Maintenance
Reliability
Availability
Maintainability
Capacity
Utility
Performance
247
t r adeof f bet ween cost s and engi neer i ng f eat ur es, and t he or gani zat i onal f unct i ons of an ent er pr i se.
Thi s i s i l l ust r at ed i n Fi gur e 1.
Har t (1985) i n a paper t i t l ed, ‘ The i nt er pr et at i on of l i f e cycl e cost s,’ r ef er s t o a pot ent i al
communi cat i on pr obl em – f i r st w i t hi n t he l i f e cycl e cost communi t y and second w i t h t he r ecei ver of t he
i nf or mat i on f or deci si on maki ng. The suggest ed sol ut i on i s t he set of def i ni t i ons at a hi gh l evel i n t er ms
of boundar i es, w hi ch consi st of (1) syst em, (2) l i f e and (3) cost . Th e syst em boundar i es est abl i sh t he
ext ent of t h e l i f e cycl e cost anal ysi s. Al t hough t hi s w oul d appear st r ai ght f or w ar d, Har t obser ves t hat
of t en anal ysi s r el at es t o t he hi ghl y vi si bl e i t ems of a syst em and di sr egar ds anci l l ar y i t ems needed t o
oper at e and mai nt ai n i t . Har t descr i bes t he boundar y of l i f e i n t he f ol l ow i ng t er ms – ‘ The l i f e of a
pr oj ect begi ns w hen t her e i s r ecogni t i on t hat a new asset i s needed t o meet t he r equi r ement s of t he
or gani zat i on. Resour ces ar e t hen exp ended by t he owner , t o t est t he sui t abi l i t y and f easi bi l i t y of t he
pr oposed asset , def i ne i t , acqui r e i t , i nt egr at e i t i nt o ser vi ce and t hen oper at e i t ’ . Har t ’ s descr i pt i on of
l i f e doesn’ t i ncl ude t he di sposal of t he syst em, w hi ch i s i mpor t ant phase of l i f e cycl e. Har t def i nes cost
as t hose r esour ces sacr i f i ced t ow ar ds an obj ect i ve.
Har t al so di scusses t he pr obl em of communi cat i ng l i f e cycl e cost i nf or mat i on t o deci si on
maker s. He concl udes t hat t her e shoul d be t w o st udi es – t he f i r st an economi c st udy and t he second a
budget ar y st udy. The economi c st udy consi der s al l t he r esour ces w hi ch w er e commi t t ed i n t he past t o
pr ocur e asset s and f r om w hi ch benef i t s can st i l l be der i ved i n addi t i on t o f ut ur e consumpt i on of
r esour ces. The budget ar y st udy concer ns onl y f ut ur e pr ocur ement and i s not concer ned w i t h t he past
(sunk) cost s.
At t he concept l evel , def i ni t i on of l i f e cycl e cost and t ot al cost of ow ner shi p i t sel f i s a chall engi ng
t ask. Def i ni t i ons of t he t er m l i f e cycl e cost i n t he l i t er at ur e ar e nor mal l y of a gener al i zed f or m. Sever al
def i ni t i ons of LCC/ TCO exi st . It i s i mpor t ant f or any or gani zat i on t o def i ne w hat t hey mean by LCC or
248
TCO. Thi s w i l l set t he boundar y f or t he cost s t hat shoul d be i ncl uded f or t he anal ysi s. I n t he f ol l ow i ng
sect i ons, w e l ook at t he some of t he most common def i ni t i ons of LCC/ TCO.
2.1 Def i ni t ions of Lif e Cycl e Cost and Cost of Ow ner shi p
Whi t e and Ost wal d (1976)
“ The l i f e cycl e cost of an i t em i s t he sum of al l f unds expended i n suppor t of t he i t em f r om i t s
concept i on and f abr i cat i on t hr ough i t s oper at i on t o t he end of i t s usef ul l i f e ”
M i chael s and Woods (1989)
“ The t ot al cost t o t he cust omer s of acqui si t i on and owner shi p of t hat syst em over i t s f ul l l i f e ”
Dhi l l on (1989)
“ The sum of al l cost s i ncur r ed dur i ng t he l i f e t i me of an i t em, i .e., t he t ot al of pr ocur ement and
ow ner shi p cost s ”
Fabr ycky and Bl anchar d (1991)
“ Al l cost s associ at ed w i t h t he syst em or pr oduct as appl i ed t o t he def i ned l i f e cycl e ”
“ Li f e cycl e cost i ng i s al l cost s associ at ed wi t h t he syst em as appl i ed t o t he def i ned l i f e cycl e. The
t ot al cost of a syst em coul d be br oken i nt o f our cat egor i es, (1) desi gn and devel opment cost , (2)
pr oduct i on/ manuf act ur i ng cost , (3) ut i l i zat i on cost , and (4) r et i r ement and di sposal cost ”
Degr aeve and Roodhoof t (1999)
“ The t ot al cost of ow ner shi p i s t he t r ue cost of buyi ng a par t i cul ar good or ser vi ce and consi st s of
pr i ce and ot her el ement s t hat r ef l ect addi t i onal cost s caused by t he suppl i er s i n t he pur chasi ng
compani es val ue chai n”
2.3 Lif e Cycl e Cost Techni que
249
Har vey (1976) i n hi s ar t i cl e compr ehensi vel y r evi ewed t he LCC t echni que and pr oposed t he gener al
pr ocedur e f or LCC, whi ch i s summar i zed i n t he Fi gur e 2.
Fi gur e 2 Har vey’ s l i f e cycl e cost i ng pr ocedur e
Woodward (1997) has elaborated the different steps of Harveys procedure as given below:
The cost elements of interest are all the cash flows that occur during the life of the asset. From the
definition of LCC it is apparent that the LCC of an asset includes all expenditure incurred in respect of it,
from acquisition until disposal at the end of its life.
Def i ni ng t he cost st r uct ur e i n vol ves gr oupi ng cost s so as t o i dent i f y pot ent i al t r adeof f s, t her eby
t o achi eve opt i mum LCC. The nat ur e of t he cost st r uct ur e def i ned wi l l depend on t he r equi r ed
dept h of t he LCC st udy, and a number of al t er nat i ve st r uct ur es have b een pr oposed i n t he l i t er at ur e
(Whi t e and Ost w al d 1976, Fabr ycky and Bl anchar d 1991).
Cost est i mat i ng r el at i onshi p i s a mat hemat i cal expr essi on t hat descr i bes, f or est i mat i ng
pur poses, t he cost of an i t em or act i vi t y as a f unct i on of one or mor e i ndependent var i abl es.
Est abl i shi ng t he met hod of LCC f or mul at i on i nvol ves choosi ng an appr opr i at e met hodol o gy t o
eval uat e t he asset ’ s LCC.
Kauf man (1970) devel op ed one of t he ear l i est f or mul at i ons of LCC wher e he has devel op ed a
model based on t he ei ght st ep appr oach i ndi cat ed bel ow and show n i n Fi gur e 3. The ei ght st eps of
Kauf man’ s LCC model ar e:
Ø Est abl i sh t he oper at i ng pr of i l e
Ø Est abl i sh t he ut i l i zat i on f act or s
Define the cost
elements of
interest
Def i ne t he
cost st r uct ur e
t o be used
Establish cost
estimating
relationships
Establish the
method of
LCC
formulation
LCC
250
Ø Ident i f y al l cost el ement s
Ø Det er mi ne al l cr i t i cal cost par amet er s
Ø Cal cul at e al l cost s at cur r ent pr i ces.
Ø Escal at e cur r ent cost s at assumed i nf l at i on r at es;
Ø Di scount al l cost s t o t he base per i od:
Ø Sum di scount ed cost s t o est abl i sh t he net pr esent val ue.
St ep 1: The oper at i ng pr of i l e (OP) d escr i bes t he per i odi c cycl e, t hr ough w hi ch equi pment w i l l go,
and i ndi cat es w hi ch equi pment w i l l , or al t er nat i vel y w i l l not be w or ki ng. The oper at i ng pr of i l e
shoul d i ndi cat e t he oper at i ng hour s of t he equi pment t hr oughout t he l i f e of t hat equi pment .
St ep 2: Ut i l i zat i on f act or s i ndi cat e i n w hat w ay equi pment w i l l be f unct i oni ng w i t hi n each mode
of t he OP.
St ep 3: Ever y cost el ement or ar ea of cost must be i dent i f i ed.
St ep 4: The cr i t i cal cost par amet er s ar e t hose f act or s, w hi ch cont r ol t he degr ee of cost s i ncur r ed
dur i ng t he l i f e of t he equi pment . St evens (1976) has suggest ed t he most si gni f i cant of t hese
ar e:
§ M ean Ti me bet ween f ai l ur es (M TBF)
§ M ean Ti me Bet w een Over haul s (M TBO)
§ M ean Ti me To Repai r (M TTR)
§ Ti me Bet ween Schedul ed M ai nt enance
§ Ener gy use r at e
St ep 5: Al l cost s ar e f i r st cal cul at ed at cur r ent r at es.
St ep 6: Al l cost s need t o be pr oj ect ed f or w ar d at appr opr i at e r at e (t hat i s, di f f er ent i al ) r at es of
i nf l at i on.
251
St ep 7: M oney has a t i me val ue and t he cash f l ow s occur r i ng i n di f f er ent t i me per i ods shoul d b e
di scount ed back t o t he base per i od t o ensur e compar abi l i t y.
St ep 8: Summi ng al l t he cash f l ows i nvol ved wi l l enabl e t he LCC of t he asset t o be est abl i shed.
Compar i sons b et w een compet i ng asset s can t hen be under t aken, and t he f al l acy of opt i n g
si mpl y f or t he asset wi t h l ow est capi t al cost wi l l t hen be exposed.
Fi gur e 3. Kauf man’ s l i f e cycl e cost i ng f or mul at i on
54. 2.4 REVIEW OF LIFE CYCLE COST MODELS
In gener al , The LCC model s can be cl assi f i ed i nt o t he f ol l owi ng cat egor i es:
1. Account i ng model s (model s t hat sum LCC component s).
2. Cost est i mat i ng r el at i onshi p (CER) model s (model s used t o anal yse desi gn al t er nat i ves).
3. Heur i st i c model s.
Parts
Labour
Parts
Operating
profile
Utilization
Factors
Maintenance
costs
PM
labour
Critical
cost para
meters,
MTBF ete
Calculate
costs at
current
prices
Escalate
current
costs
(inflation)
Discount
costs to
base
period
Sum up all
the cash
flows
involved
LCC
CM
labour
Operating
manpower
Electricity,
water etc.
Operating
costs
Initial
spares cost
Initial
acquisition
cost
Overhaul
cost
Parts
Labour
Parts
Operating
profile
Utilization
Factors
Maintenance
costs
PM
labour
Critical
cost para
meters,
MTBF ete
Calculate
costs at
current
prices
Escalate
current
costs
(inflation)
Discount
costs to
base
period
Sum up all
the cash
flows
involved
LCC
CM
labour
Operating
manpower
Electricity,
water etc.
Operating
costs
Initial
spares cost
Initial
acquisition
cost
Overhaul
cost
Operating
profile
Utilization
Factors
Maintenance
costs
PM
labour
Critical
cost para
meters,
MTBF ete
Calculate
costs at
current
prices
Escalate
current
costs
(inflation)
Discount
costs to
base
period
Sum up all
the cash
flows
involved
LCC LCC
CM
labour
Operating
manpower
Electricity,
water etc.
Operating
costs
Initial
spares cost
Initial
acquisition
cost
Overhaul
cost
252
4. Fai l ur e f r ee war r ant y model s (model s used t o anal yse war r ant y per i ods)
5. Rel i abi l i t y model s (used t o appor t i on r el i abi l i t y and mai nt ai nabi l i t y), and
6. Economi c anal ysi s model s (model s deal i ng wi t h gener al cost ef f ect i veness).
How ever , Sher i f and Kol ar i k (1981) cl assi f y LCC model s i nt o t hr ee gener al f or ms: (1) concept ual , (2)
anal yt i cal model s, and (3) heur i st i c model s. Concept ual model s consi st of a set of hypot het i cal
r el at i onshi ps expr essed i n a qual i t at i ve f r amewor k. Concept ual model s ar e gener al l y const r uct ed at
macr o l evel . Anal yt i c model s consi st of a set of mat hemat i cal r el at i onshi ps, whi ch ar e used t o descr i be a
cer t ai n aspect of t he syst em. Such model s r ange f r om model s cover i ng ver y speci f i c aspect s of a syst em
t o model s, w hi ch addr ess t ot al syst em l i f e cycl e cost .
Gupt a (1983) i dent i f i es t hr ee t ypes of anal yt i c model s: (1) desi gn t r adeof f model s, (2) t ot al cost
model s, and (3) l ogi st i c suppor t model s. Desi gn t r adeof f model s r el at e t o t he desi gn phase of t he l i f e
cycl e cost and at t empt t o mi ni mi se cost t o meet a gi ven val u e of desi gn par amet er s such as r el i abi l i t y
and avai l abi l i t y t o maxi mi ze t he val u e of desi gn par amet er s f or gi ven cost const r ai nt s. Tot al cost model s
ar e t er med t r ue l i f e cycl e cost model s and usual l y encompass t he t ot al l i f e of t he syst em. They at t empt
t o mi ni mi ze t he t ot al l i f e cycl e cost of t he syst em w hi le maxi mi zi ng i t s p er f or mance and ef f ect i veness by
eval uat i ng var i ous par amet er s such as r el i abi l i t y, mai nt ai nabi l i t y, avai l abi l i t y et c, w hi ch af f ect l i f e cycl e
cost . Logi st i c suppor t model s ar e concer ned w i t h t he oper at i ons phase of t he l i f e cycl e. Usual l y t he
obj ect i ve of su ch model s i s t o det er mi n e cost s f or al t er nat i ve sup por t pl ans and ef f ect on t he syst em’ s
ef f ect i veness. They r ef l ect oper at i ons cost par amet er s as var i abl e cost s and r esear ch, devel opment ,
t est and eval uat i on and acqui si t i ons cost s as f i xed cost s. These model s ar e i nconsi st ent i n t hat desi gn
par amet er s such as r el i abi l i t y and mai nt ai nabi l i t y heavi l y i nf l uence oper at i ons cost s and t her ef or e f al l
shor t of det er mi ni ng opt i mal l i f e cycl e cost .
Dhi l l on (1989) si mpl y di vi des l i f e cycl e cost model s i nt o t w o f or ms: (1) gener al l i f e cycl e cost model s,
and (2) speci f i c l i f e cycl e cost model s. Gener al l i f e cycl e cost model s ar e not r el at ed t o any sp eci f i c
253
equi pment or syst em w her eas speci f i c l i f e cycl e cost model s have been d evel oped f or par t i cul ar t ypes of
equi pment or syst em. Gi ven t he speci f i c i nt er r el at i o nshi ps and i nt er act i ons of a par t i cul ar syst em, t he
appl i cat i on of gener al model s i s cl ear l y l i mi t ed. Recent l y, Dani el (1991) has cl assi f i ed l i f e cycl e cost
model s i nt o t w o br oad cat egor i es: (1) account i ng model s w hi ch at t empt t o assembl e and di st r i but e
cost s, det er mi ned el sewher e so as t o descr i be t he t ot al cost of a syst em, and (2) pr edi ct i ve model s
w hi ch ar e used t o f or ecast t he val ues of t he var i ous cost el ement s r equi r ed as i nput t o t he account i ng
model s.
Nobl e and Tanchoco (1990) devel oped a concept ual f r amew or k f or concur r ent desi gn and economi c
j ust i f i cat i on of t he syst em. A pr ot ot ype i mpl ement at i on w as devel oped t o expl or e t he usef ul ness of
t he desi gn j ust i f i cat i on concept . Act ual dat a f r om t he desi gn of an el ect r omagnet i c/ r adi o f r equency
shi el d, a component i n el ect r i cal met er i ng equi pment , was used t o demonst r at e t he model .
Woodw ar d (1997) i n hi s paper t i t l ed, ‘ Li f e Cycl e Cost i ng – Theor y, Inf or mat i on Acqui si t i on and
Appl i cat i on,’ pr esent ed a case on t ot al cost of own er shi p on Sout h Yor kshi r e Passenger Tr anspor t
(SYPT). SYPT’ s mai n act i vi t y i s t he pr ovi si on of passenger t r anspor t ser vi ces b y r oad. It ’ s f i xed asset s
w er e w or t h $ 43,327,500 out of w hi ch t he passenger vehi cl es account ed f or about 17,662,500. The
company pur chases vehi cl es t hat f or m a maj or par t of t he capi t al expendi t ur e, on a r egul ar basi s
and t he deci si on t o pur chase t hem i s based on t he LCC t echni que. The est i mat ed l i f e cycl e cost s ar e
di scount ed at an assumed monet ar y cost of capi t al of 15%, af t er i ncl udi ng a st andar d i nf l at i on r at e
assumed over t he l i f e of t he asset . If t he t w o al t er nat i ves have si mi l ar di scount ed cost s, t hen a
choi ce w i l l be made by t he f i nanci al di r ect or t aki ng i nt o account nonf i nanci al f act or s such as t h e
cr edi bi l i t y, r el i abi l i t y et c. of t he suppl i er s. Al t hough, t he case was on passenger t r anspor t by r oad,
t he concept i s val i d f or any syst em, i ncl udi ng ai r bor ne def ence equi pment .
Degr aeve and Roodhoof t (1999) devel op ed a mat hemat i cal pr ogr ammi ng model t hat uses t ot al cost
of owner shi p i nf or mat i on t o sel ect suppl i er s and det er mi n e or der quant i t i es over a mul t i p er i od t i me
254
hor i zon. The t ot al cost of ow ner shi p quant i f i es al l cost s associ at ed w i t h pur chasi ng pr ocess and i s based
on t he act i vi t i es and cost dr i ver s det er mi ned by an Act i vi t y Based Cost i ng (ABC) syst em. They have al so
di scussed a case on t he pur chasi ng pr obl em of heat i ng el ect r odes at Cocker i l l Sambr e, a Bel gi an
mul t i nat i onal st eel pr oducer . In t hi s case, qual i t y i ssues account ed f or mor e t han 70% of t he t ot al cost
of ow ner shi p maki ng t he qual i t y of t he suppl i er a cr i t i cal success f act or i n t he suppli er sel ect i on pr ocess.
LCC model can be a si mpl e ser i es of cost est i mat i on r el at i onshi ps (CERs). LCC anal ysi s dur i ng t he
concept ual or pr el i mi nar y desi gn phases may r equi r e t he use of basi c account i ng t echni ques (Fabr ycky
and Bl anchar d, 1991). The most i mpor t ant t ask i n LCC model i ng i s t he const r uct i on of Cost Br eakdow n
St r uct ur e (CBS), w hi ch show s var i ous cost cat egor i es t hat combi ne t o pr ovi de t he t ot al cost . Cost
br eakdow n st r uct ur e shoul d exhi bi t t he f ol l ow i ng basic char act er i st i cs (Bl anchar d et al 1995):
1. Al l syst em cost el ement s must be consi der ed.
2. Cost cat egor i es ar e gener al l y i dent i f i ed wi t h a si gni f i cant l evel of act i vi t y or some maj or i t em of
har dwar e.
3. The cost st r uct ur e and cat egor i es shoul d be coded i n such a manner as t o al l ow f or t he anal ysi s
of cer t ai n speci f i c ar eas of i nt er est (e. g., syst em oper at i on, ener gy consumpt i on, equi pment
desi gn, spar es, mai nt enance per sonnel and suppor t , mai nt enance equi pment and f aci l i t i es). In
some i nst ances, t he anal yst may w i sh t o pur sue a desi gnat ed ar ea i n d ept h w hi l e cover i ng ot her
ar eas wi t h gr oss t opl evel est i mat es. Thi s wi l l cer t ai nl y occur f r om t i me t o t i me as a syst em
evol ves t hr ough t he di f f er ent phases of i t s l i f e cycl e.
4. When r el at ed t o a speci f i c pr ogr am, t he cost st r uct ur e shoul d be compat i bl e (t hr ough cr oss
i ndexi ng, codi ng et c.) w i t h t he cont r act w or k br eakdow n st r uct ur e (WBS) and w i t h management
account i n g pr ocedur es used i n col l ect i ng cost s.
5. For pr ogr am, wher e subcont r act i ng i s pr eval ent , i t i s of t en desi r abl e and necessar y t o separ at e
suppl i er cost s (i .e., i ni t i al bi d pr i ce and f ol l ow on pr ogr am cost s) f r om ot her cost s. The cost
255
st r uct ur e shoul d al l ow f or t he i dent i f i cat i on of sp eci f i c w or k packages t hat r equi r e cl ose
moni t or i ng and cont r ol .
An exampl e of a cost br eakdow n st r uct ur e adopt ed f r om Bl anchar d (1991) i s show n i n Fi gur e 4.
Ref er r i ng t o Fi gur e 4, cost s may be accumul at ed at di f f er ent l evel s dependi ng on t he ar eas of i nt er est
and t he dept h of det ai l r equi r ed. M ost of t he LCC model s can be a si mpl e ser i es of cost est i mat i on
r el at i onshi ps. Est i mat i ng model s used i n i ndust r y can be br oadl y cl assi f i ed as par amet r i c model s,
anal ogous model s and det ai l ed model s (Asi edu and Gu, 1998).
256
Fi gur e 4. Cost Br eakdow n St r uct ur e (Bl anchar d, 1991)
Par amet r i c model s i nvol ve gener at i on and appl i cat i on of equat i ons t hat descr i be r el at i onshi ps
bet ween cost schedul es and measur abl e at t r i but es of a syst em t hat must be br ou ght f or t h, sust ai ned
and r et i r ed (Dean, 1995). Cost est i mat i on w i t h a par amet r i c model i s based on pr edi ct i ng a pr oduct ’ s
cost ei t her i n t ot al or f or var i ous act i vi t i es, by t he use of r egr essi on anal ysi s based on hi st or i cal cost and
t echni cal i nf or mat i on. A si mpl e par amet r i c CER i s t he r el at i on bet w een t he cost of bui l di ngs and t he
f l oor ar ea. M ost of t he cost est i mat i ng r el at i onshi ps f or ai r bor ne mi l i t ar y syst ems r el at es t he cost t o t he
par amet er s such as wei ght , cr ui se speed, et c. of t he syst em. Par amet r i c est i mat i ng can i nvol ve
consi der abl e ef f or t because of t he syst emat i c col l ect i on and r evi si on pr ocess r equi r ed t o keep t he CERs
updat ed, but once t hi s dat a i s avai l abl e est i mat es can be pr oduced f ai r l y r api dl y (Gr eves and Schr ei ber ,
1993). Ther e ar e sever al commer ci al model s avai l abl e now. The most wi del y used i s t he Lockheed
M ar t i n’ s PRICE syst em. Est abl i shment s such as Br i t i sh Aer ospace, The Eur opean Space Agency and
Total System Cost C
T
Research and Development, C
R
Investment, C
I
Operation and Maintenance, C
O
Program Management, C
RM
Advanced R & D, C
RR
Engineering Design, C
RE
Engineering Development
and Test, C
RT
Engineering Data, C
RD
Manufacturing, C
IM
Construction, C
IC

Manufacturing Facilities
Test Facilities
Operational Facilities
Maintenance Facilities
Manufacturing, C
IM
Operations, C
OO
Maintenance, C
OM
System/Equipment
Modification, C
ON
System Phaseout and
Disposal, C
OP
Total System Cost C
T
Research and Development, C
R
Investment, C
I
Operation and Maintenance, C
O
Program Management, C
RM
Advanced R & D, C
RR
Engineering Design, C
RE
Engineering Development
and Test, C
RT
Engineering Data, C
RD
Manufacturing, C
IM
Construction, C
IC

Manufacturing Facilities
Test Facilities
Operational Facilities
Maintenance Facilities
Manufacturing, C
IM
Operations, C
OO
Maintenance, C
OM
System/Equipment
Modification, C
ON
System Phaseout and
Disposal, C
OP
257
NASA use PRICE syst em. How ever , i t i s not r ecommended f or est i mat i ng t he cost of pr oduct s t hat ut i l i ze
new t echnol ogi es.
55. ANALOGOUS MODELS
Cost est i mat i ng made by anal ogy i dent i f i es a si mi l ar pr oduct or component and adj ust s f or
di f f er ences bet w een i t and t he t ar get pr oduct (Shi el ds and Young 1991). The ef f ect i veness of t hi s
met hod depends heavi l y on an abi l i t y t o i dent i f y cor r ect l y t he di f f er ences b et w een t he case i n hand and
t hose deemed t o be compar abl e. The mai n di sadvant age of est i mat i ng by anal ogy i s t he hi gh degr ee of
j udgment r equi r ed.
56. DETAILED MODELS
Det ai l ed model s use est i mat es of l abour t i mes and r at es and al so mat er i al quant i t i es and pr i ces
t o est i mat e t he di r ect cost s of a pr oduct or act i vi t y (Shi el ds and Young, 1991). An al l ocat i on r at e i s t hen
used t o al l ow f or i ndi r ect / o ver head cost s. Thi s i s know n as bot t omup est i mat i ng and i s w i del y used t o
al l ow i ndi r ect / over head cost s. It i s t he most t i me consumi ng and cost l y appr oach and r equi r es a ver y
det ai l ed kn owl edge of t he pr oduct and pr ocesses. However , t he most accur at e cost est i mat es can be
made usi ng t hi s appr oach. The met hod i nvol ves (Asi edu and Gu, 1998) est i mat i o n t he t i me needed
t o per f or m an act i vi t y and t he hour l y r at es f or t he man and machi ne, and t hen mul t i pl y t i mes and r at es
t o get cost s. Ti me st andar ds can be i ndust r y st andar ds, i nhouse st andar ds or based on exper t guesses.
Inhouse st andar ds ar e t he best but most di f f i cul t t o devel op. Indust r i al t i me st andar ds f or pr oduct i on
oper at i ons exi st f or many common t asks.
258
57. IN THE NEXT FEW SECTIONS WE DESCRIBE FEW LCC
MODELS THAT ARE POPULAR AMONG PRACTITIONERS THAT CAN
BE USED FOR ESTIMATION OF LIFE CYCLE COST/COST OF
OWNERSHIP OF AIRBORNE MILITARY EQUIPMENT.
58. 2.5 TAYLOR`S LCC MODEL
Tayl or ’ s model f ocuses on t he capi t al and r evenue cost s. Tayl or cl ai ms t hat i n any di scussi on of
t r adeof f s bet w een i ni t i al and subsequent cost s, a point t hat i s f r equent l y made i s t hat t her e i s a maj or
di st i nct i on bet w een i ni t i al capi t al cost s and r evenue cost s. It i s cl ai med t hat compani es and publ i c
bodi es f aced w i t h l i mi t ed capi t al budget or cost l i mi t s do not have t he f aci l i t y t o i ncr ease i ni t i al capi t al
cost s on t he chance t hat t her e w i l l be f ut ur e r evenue gai ns. How ever , Tayl or cl ai ms t hat t he di st i nct i on
bet w een r evenue exp endi t ur e and capi t al i s an account i ng one whi ch doesn’ t af f ect t he l i f e cycl e cost
concept based on t he cash f l ow s t hr oughout t he l i f e of t he asset .
259
Fi gur e 5. Tayl or ’ s LCC cost el ement s and i nt er act i on
Tayl or ’ s cost s of ow ni ng physi cal asset ar e show n i n Fi gur e 5. The cost s f al l i nt o t hr ee gr oups,
f i r st t he i ni t i al capi t al cost s secondl y t he r evenue cost s of oper at i ng and mai nt ai ni ng t he asset dur i ng i t s
oper at i onal l i f e and t hi r dl y t he cost of asset di sposal , w hi ch may be r evenue of capi t al i f i t i s subst ant i al .
The i ni t i al cost s f or an or gani zat i on w hi ch desi gns and const r uct s physi cal asset s f or i t s ow n use or f or
r esal e woul d be:
Ø Resear ch and Devel opment
Ø Desi gn and Speci f i cat i on.
Ø M anuf act ur i ng.
Ø Qual i t y cont r ol and t est i ng.
Ø M oni t or i ng Per f or mance.
Capital
Costs
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
;
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
'
¹
etc Tools Space, inventory, spares, of provision A
e Maintenanc  Training and Manuals
Operations  Training and Manuals
ioning on/Commiss Installati
e/Build Manufactur
t Developmen
Design
ion Specificat
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
;
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
'
¹
Down time 
Overheads ent Establishm 
Equipment and Facilities 
Labour 
Spares  Costs e Maintenanc
Labour and material Indirect 
Overheads
and Expenses Direct 
Labour Direct 
Material Direct  Costs Operating
Revenue
Costs
Capital and
Revenue Residual
Costs
¹
¹
¹
¹
¹
;
¹
¹
¹
¹
¹
¹
'
¹
Disposal 
n Dislocatio 
Demolition  Costs Disposal
Value Disposal
Residual Costs
Preventive
Repair
Output Quantity
Output Quality
Material Utilization
Labour Utilization
Asset Utilization
Asset Reliability
Asset Maintainability
Asset Availability
Capital
Costs
Capital
Costs
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
;
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
'
¹
etc Tools Space, inventory, spares, of provision A
e Maintenanc  Training and Manuals
Operations  Training and Manuals
ioning on/Commiss Installati
e/Build Manufactur
t Developmen
Design
ion Specificat
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
;
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
¹
'
¹
Down time 
Overheads ent Establishm 
Equipment and Facilities 
Labour 
Spares  Costs e Maintenanc
Labour and material Indirect 
Overheads
and Expenses Direct 
Labour Direct 
Material Direct  Costs Operating
Revenue
Costs
Revenue
Costs
Capital and
Revenue Residual
Costs
Capital and
Revenue Residual
Costs
¹
¹
¹
¹
¹
;
¹
¹
¹
¹
¹
¹
'
¹
Disposal 
n Dislocatio 
Demolition  Costs Disposal
Value Disposal
Residual Costs
Preventive
Repair
Output Quantity
Output Quality
Material Utilization
Labour Utilization
Asset Utilization
Asset Reliability
Asset Maintainability
Asset Availability
260
The second gr oup of cost s ar e i ncur r ed dur i ng t he oper at i onal l i f e of t he asset and t hi s w oul d
i ncl ude t he cost s of : oper at i ng t he asset s i ncl udi ng t he l abour , mat er i al s, t ool s, f i xt ur es and over heads,
mai nt enance i ncl udi ng spar es and l abour . Fi nal l y t her e ar e di sposal cost s w hi ch i ncl ude cost s of
demol i t i on and r emoval , di sl ocat i on of exi st i ng pr oduct i on capaci t y. Agai nst t hi s may be any di sposal
val ue of t he physi cal asset .
2.6 Raymer ’ s LCC M o del
Raymer ’ s l i f e cycl e cost i ng model i s based on t he Devel opment and Pr ocur ement Cost s of
Ai r cr af t (DAPCA IV) model devel oped b y t he Rand Cor por at i on. The Rand cor por at i on devel oped sever al
cost est i mat i on r el at i onshi ps f or est i mat i ng var i ous cost s f or al l depar t ment s i ncl udi ng engi neer i n g,
t ool i ng, manuf act ur i ng and qual i t y cont r ol gr oups. DAPCA assumes a t enyear pr oduct l i f e, w hi ch al so i s
an i ndust r y st andar d. Rand Cor por at i on cl ai ms t hat DAPCA, coupl ed w i t h appr opr i at e f act or s i s accur at e
t o w i t hi n +/  5% of act ual cost s.
2.7 Roskam LCC M o del
Roskam model di vi des t he LCC i nt o f our maj or cat egor i es: (1) Resear ch and Devel op ment , t est and
eval uat i on, (2) Pr ogr am acqui si t i on cost t hat i ncl ud es manuf act ur i n g cost and manuf act ur er ’ s pr of i t , (3)
Oper at i ng cost , and (4) Di sposal cost . The expr essi on f or LCC i s gi ven by t he equat i on:
D OM AC RDTE
C C C C LCC + + + ·
(1)
Wher e, C
RDTC
i s t he R& D, t est and eval uat i on cost , C
AC
i s t he acqui si t i on cost , C
OM
i s t he oper at i on
and mai nt enance cost and C
D
i s t he di sposal cost .
261
R& D cost i s f ur t her br oken i nt o cost el ement s such as: (1) ai r f r ame engi neer i ng and desi gn, (2) t est
f l i ght ai r cr af t and f li ght t est oper at i ons, (3) t est & si mul at i on f aci l i t i es, (4) cost t o f i nance. Each of t hese
cost el ement s i s est i mat ed usi ng par amet r i c met hods usi ng ai r cr af t w ei ght , maxi mum desi gn speed, and
number of ai r cr af t bui l t . Si mil ar l y acqui si t i on cost i s cal cul at ed usi ng par amet er s such as t he number of
ai r cr af t s manuf act ur ed, manuf act ur i ng cost , t akeof f wei ght , desi gn cr ui se speed et c.
Oper at i on cost s ar e br oken i nt o t he mat er i al cost s, di r ect and i ndi r ect p er sonnel cost and l ogi st i c
suppor t cost s. The di sposal cost i s t aken as 1% of t he LCC cost . Roskam devel op ed sever al cost
est i mat i on r el at i onshi ps f or est i mat i on of t he var i ous cost s gi ven i n above equat i on. M ost of t hese
model s wer e devel oped usi ng wei ght of t he ai r cr af t as a dependent var i abl e.
2.8 Fabr ycky and Bl anchar d’ s LCC M odel
Fabr ycky and Bl anchar d (1991) devel op ed t he det ai l ed LCC model . The most i mpor t ant t ask i n
t hei r model i s t o d evel op t he cost br eakdow n st r uct ur e (CBS, show n i n Fi gur e 4). Ther e i s no met hod set
f or br eaki ng dow n t h e cost s as l ong as t he met hod used can be t ai l or ed t o t he speci f i c appl i cat i on.
Pr i mar i l y t he cost i s di vi ded i nt o t he f ol l owi ng f our cat egor i es:
Ø Resear ch and devel opment
Ø Pr oduct i on and const r uct i on cost s
Ø Oper at i on and mai nt enance cost s
Ø Ret i r ement and di sposal cost s
Thus t he t ot al cost i s (C) i s cal cul at ed usi ng t he expr essi on:
D O P R
C C C C C + + + ·
(2)
Wher e,
262
Cost Disposal and Retirement C Cost, e Maintenanc and Operation C
Cost, on Constructi and Production C cost, D & R C
R O
P R
· ·
· ·
The t ot al cost , C, i ncl udes al l f ut ur e cost s associ at ed w i t h t he acqui si t i on, ut i l i zat i on, and
subsequent di sposal of syst em equi pment .
Resear ch and d evel opment cost i ncl udes al l cost s associ at ed wi t h concept ual f easi bi l i t y st udi es,
basi c and advanced r esear ch and devel op ment , engi n eer i ng desi gn, f abr i cat i on and t est of engi neer i ng
pr ot ot ype model s (har dwar e), and associ at ed document at i on. Al so cover s al l r el at ed pr ogr am
management f unct i ons. The R& D cost i s gi ven by:
RD RT RE RR RM
C C C C C + + + + ·
R
C
(3)
Wher e,
cost data g Engineerin C
cost t/test developmen g Engineerin C cost, design g Engineerin C
Cost, D & R Advanced C cost, management Program C
RD
RT RE
RR RM
·
· ·
· ·
Oper at i ons and mai nt enance cost i ncl udes al l cost s associ at ed w i t h t he oper at i on and
mai nt enance suppor t of t he syst em t hr oughout t he l i f e cycl e subsequent t o t he equi pment del i ver y i n
t he f i el d. Speci f i c cat egor i es cover t he cost of syst em oper at i on, mai nt enance, sust ai ni ng l ogi st i c
suppor t , equi pment modi f i cat i ons. Thus, t he oper at i on and mai nt enance cost i s gi ven by:
OP ON OM OO O
C C C C C + + + ·
(4)
Wher e,
disposal system of Cost C ons, modificati cycle life system of Cost C
e, maintenanc cycle life system of Cost C , operations cycle life system of Cost C
OP ON
OM OO
· ·
· ·
263
The cost s i n equat i ons (2) – (4) can be f ur t her di vi ded i nt o var i ous cost el ement s.
2.9 Bur ns Lif e Cycl e Cost M o del
Bur ns devel op ed a cost est i mat i on r el at i onshi p f or pr edi ct i ng l i f e cycl e cost of ai r cr af t based on i t s
wei ght . Bur ns model i s a si mpl e ext ensi on of Roskam’ s l i f e cycl e cost model . The model al so i ncl udes
j udgement f act or f or comput i n g ai r f r ameengi neer i ng hour s f or devel opment and pr oduct i on. A
compl et e anal ysi s of Bur ns model i s pr esent ed i n Jayakr i shnan (2002).
2.10 PRICE Lif e Cycl e Cost i ng Syst em
The PRICE syst em consi st s of par amet r i c cost est i mat i on model s f or pr edi ct i ng t he l i f e cycl e cost of
weapon syst ems devel oped by t he Lockh eed M ar t i n. The PRICE syst em’ s t ool i ncl udes a set of f our
par amet r i c cost est i mat i on model s, each w i t h a di f f er ent speci al t y ar ea. The model s consi st s of :
PRICE M: Thi s model speci f i cal l y addr esses el ect r o ni c modul e l evel har dwar e devel opment and
pr oduct i on cost s.
PRICE H: Thi s model speci f i cal l y addr esses t he cost s associ at ed w i t h d evel opment and pr oduct i on of
har dw ar e. Thi s t ool can use out put s of t he PRICE M t ool .
PRICE HL: Thi s model uses dat a gener at ed by PRICE H and cal cul at es t he har dw ar e l i f ecycl e cost s,
i ncl udi ng spar i ng f or a depl oyment envi r onment .
PRICE Sof t ware: Thi s model est i mat es bot h devel opment cost s and l i f e cycl e suppor t cost s f or
sof t war e.
2.11 Equi pment Desi gner ’ s Cost Analysi s Syst em (EDCAS) M odel
EDCAS i s one of t he popul ar commer ci al syst ems avai l abl e f or l i f e cycl e cost pr edi ct i on. EDCAS i s a
sequent i al model and i s appl i cabl e f or desi gn t o LCC i n f r ont end desi gn anal ysi s. Over 500 gover nment
264
and i ndust r y use t he syst em w or l dw i de. For exampl e, U. S. Ai r For ce uses EDCAS f or ai r cr af t and
ai r bor ne weapons and el ect r oni c syst ems.
2.12 LCC M odel s Usi ng M ar kov Chai n
St ump (1988) devel op ed a LCC model based on M ar kov chai ns and i l l ust r at ed t he model f or a
hypot het i cal r emot el y pi l ot ed vehi cl e (RPV). The M ar kov chai n i s used t o est i mat e t he oper at i on,
mai nt enan ce and suppor t cost s. The model assumes t hat t he syst em goes t hr ough a number of st at es.
For any st at e, t he number of vi si t s per cycl e mul t i pl i ed by t he cost per vi si t and t he expect ed l i f e of t he
RPV i n cycl es w i l l yi el d a l i f e cost f or t hat st at e. Summi ng over al l st at es w i l l yi el d a t ot al l i f e cost . To
f ul l y i mpl ement t hi s l i f e cycl e cost met hodol ogy, t he f ol l owi ng i nf or mat i on i s needed:
1. A l i st of syst em st at es.
2. A l i st of t r ansi t i on pr obabi l i t i es f r om any st at e t o any ot her st at e (zer o i f t he st at es do not
communi cat e).
3. A l i st of t he cost s of ent er i ng t he st at es.
4. The aver age number of vi si t s per cycl e f or each st at e.
5. The expect ed l i f e of t he syst em.
6. Cost est i mat i ng r el at i onshi ps f or comput i ng cost s.
The l i f e cycl e cost f or st at e i i s:
i i i
C a L LC × × ·
(5)
Wher e, a
i
, i s t he st at e pr obabi l i t y f or st at e i and C
i
i s t he aver age cost of ent r y i nt o st at e i and L i s
t he expect ed l i f e of t he syst em. The t ot al l i f e cost of t he RPV i s si mpl y t he sum of t he LC
i
f or al l st at es.
One of t he maj or pr obl ems w i t h M ar kov chai n model i s t hat t he syst em i s l i kel y t o have l ar ge number of
st at es.
265
2.13 LCC model f or l abour f act or
Dahl en and Bol msj o (1996) devel op ed a l i f e cycl e cost model f or t he l abour f act or t hat cover s
t he cost s f or an empl o yee over t he whol e empl oyment cycl e – f r om t he r equi r ement unt i l r et i r ement .
The cost s ar e di vi ded i nt o t hr ee basi c cat egor i es:
1. Empl oyment cost s: consi st i ng of cost s f or r ecr ui t i ng, i nt r oduct i on and t r ai ni ng of new empl oyees
– t o compar e w i t h acqui si t i on cost s such as pr oj ect i ng, i nst al l at i on and st ar t up of t he new
equi pment .
2. Oper at i ons cost s: consi st i ng of w ages, and l abour r el at ed over head – t o compar e w i t h
depr eci at i on, mai nt enance and r epai r s.
3. Wor k envi r on ment al cost s: consi st i ng of a addi t i onal cost s f or absent eei sm, r ehabi l i t at i on and
pensi ons – t o compar e w i t h cost s f or i ncr eased mai nt enance and r epai r s and f i nal l y t o scr ap t he
equi pment .
The basi c cat egor i es of l abour l i f e cycl e cost can be di vi ded i nt o: empl o yment cost s, oper at i on cost s
and wor k envi r onment al cost s. The empl o yment cost s can be di vi ded i nt o t hr ee maj or sub cat egor i es:
(1) r ecr ui t ment cost s, (2) addi t i onal pr oduct i ons cost s and (3) educat i on cost s.
Oper at i on cost s ar e i ncur r ed w hen t he empl oyee i s i n t r oduced and mast er s t he w or k t asks, t he
cost s consi st s of w ages and over heads. The t hi r d cat egor y, w or k envi r onment al cost s, i ncl udes cost s f or
absence, si ckness benef i t s, r ehabi l i t at i on cost s and di sabi l i t y pensi on cost s.
2.14 LCC M odel s i n Desi gni ng f or Logi st i c Suppor t
Hat ch and Bedi nel l i (1999) devel oped a model t hat car r i es out a concur r ent opt i mi zat i on of a
pr oduct desi gn and i t s associ at ed manuf act ur i ng and l ogi st i c suppor t syst em. The model i s const r uct ed
266
w hi ch l i nks t oget her t he deci si ons associ at ed w i t h t hr ee maj or phases of t he l i f e cycl e: pr oduct desi gn,
manuf act ur i ng and l ogi st i c syst em desi gn, pr oduct i on and f i el d oper at i on cont r ol . The model i ncl uded
an opt i mi zat i on scheme t hat concur r ent l y opt i mi zes t he deci si on var i abl e of t he l i nked model . The f i nal
sol ut i on pr escr i bed by t he model i s based on a mul t i cr i t er i a val ue f unct i on f or med f r o m t he i ndi vi dual
obj ect i ves of mi ni mi zi ng l i f e cycl e cost and maxi mi zi ng avai l abi l i t y. The model eval uat es al t er nat i ve
desi gn sol ut i ons by cal cul at i ng t he associ at ed oper at i onal avai l abi l i t y as w el l as manuf act ur i ng and
l ogi st i c suppor t cost s. The t wo mai n per f or mance measur es can be combi ned i nt o t h e f ol l ow i ng bi 
cr i t er i a model f or mul at i on:
Mi n Li f e Cycl e Cost
Max Syst em Avai l abi l i t y
Subj ect t o Pr oduct Desi gn Requi r ement s
2.15 Appl i cat i ons of LCC/ Cost of Ow ner shi p M odel s
The l i t er at ur e sur vey car r i ed out by Ni chol as (1999) i ndi cat es t hat t he t er m l i f ecycl e cost / cost of
ow ner shi p i s appl i ed t o var yi n g f or ms of anal ysi s, w hi ch ar e under t aken f or a r ange of di f f er ent
pur poses. These appl i cat i ons can be br oadl y descr i bed as eval uat i on and deci si onmaki ng, pl anni ng and
budget i ng, cost management and cont r ol , pr oj ect / pr ogr am management and cont r ol , l i f ecycl e
management , cont r act i ng, and mar ket i ng. These appl i cat i ons ar e show n i n Tabl e 1 i n r el at i on t o r ecent l y
publ i shed paper s.
Al t hough l i f ecycl e cost / cost of ow ner shi p i s appl i ed t o anal ysi s, w hi ch i s used f or a w i de r ange of
pur poses, i t can be ar gued t hat i n essence t he anal ysi s i s bei ng used t o pr ovi de i nf or mat i on f or t w o
f undament al pur poses  pl anni ng and cont r ol . Pl anni ng i ncl udes i ) d eci si onmaki ng  t he al l ocat i on of
267
r esour ces f or f ut ur e per i ods t hr ough t he i dent i f i cat i on, eval uat i on and sel ect i on of al t er nat i ve cour ses
of act i on and ii ) budget i ng  t he i dent i f i cat i on of means r equi r ed t o i mpl ement t he sel ect ed cour ses of
act i on. Pr i nci pal ar eas of cont r ol i ncl ude cost cont r ol and cont r act ual ar r angement s. Cost cont r ol
i ncl udes pr ocedur es t o i nf l uence cost t hr ough desi gn (M i chael s and Wood, 1989) and t hr oughout t h e
pr ocess of acqui si t i on (US Depar t ment of Def ense, 1996). Cont r act ual ar r angement s ar e desi gned t o
cont r ol cost t hr ough l egal agr eement s. Cont r act ual ar r angement s i ncl ude guar ant ees f or par t of l i f e
cycl e cost such as suppor t cost (Baat he, 1995) and guar ant ees f or t ot al l i f e cycl e cost (Aksel sson and
Bur st r om, 1994). Pl anni ng and cont r ol ar e i nt egr at ed t hr ough management w hi ch i ncl udes pr oj ect and
pr ogr am management dur i ng acqui si t i on and much wi der l i f ecycl e management whi ch ext ends t o t he
compl et e l i f ecycl e of t he syst em.
Ther e i s an i mpor t ant di st i nct i on t o be made bet w een t he nat ur e of t he use of anal ysi s t o pr ovi de
i nf or mat i on f or pl anni ng and cont r ol . In pl anni ng, i n bot h deci si onmaki ng and budget i ng, l i f ecycl e cost
i s used as an ex ant e concept t o pr edi ct f ut ur e cost . In cont r ol , i t i s used as bot h an ex ant e and an ex
post concept . It i s used as an ex ant e concept t o est abl i sh t ar get s or per f or mance cr i t er i a but as an ex
post concept w hen moni t or i ng and compar i ng cost per f or mance i n t er ms of pl anned cost agai nst act ual
cost . As i ndi cat ed i n t he di scussi on of concept s of cost bel ow , t hi s w i l l i nvol ve t he appl i cat i on of
di f f er ent concept s of cost .
Anal ysi s usi ng t he concept of l i f ecycl e cost i s not onl y used f or di f f er ent pur poses but al so i nvol ves
di f f er ent f unct i onal user s. Exampl es of di f f er ent user s i ncl ude pol i cymaker s (Ki r kpat r i ck, 1996),
manager s (Ri ggs and Jones, 1990, Gr eene, 1991) and engi n eer s (Var i ous  see r ef er ences and
bi bl i ogr aphy). Pol i cymaker s ar e i nvol ved i n st r at egi c deci si ons t hat i nvol ve t he l ongt er m
commi t ment of f unds. M anager s may be pr oj ect manager s or budget manager s whose i nt er est i s
pr i nci pal l y i n cont r ol . Engi neer s may i ncl ude desi gn engi neer s, pr oduct i on engi neer s, syst ems
268
engi n eer s, l ogi st i cs engi n eer s and ot her s who have an i nt er est i n l i f ecycl e cost f or engi neer i ng
deci si onmaki ng.
Tabl e 1. Appl i cat i on of LCC/ TCO i n t he l i t er at ur e
Appl i cat i on of Li f e Cycl e Cost and Cost of Ownershi p Model s
Desi gn eval uat i on Dacko and Dar l i ngt on (1988) Takagi shi (1989)
Gi bbs and Ki ng (1989) Johnson (1990) Keene and
Keene (1993) St ahl and Wal l ace (1995) Pl ebani ,
Rosi and Zanet t a (1996) Asi edu and Gu (1998)
M at er i al s sel ect i o n Wi nkel (1996)
Choi ce of desi gn l i f e Howar d (1991) Asi edu and Gu (1998)
Envi r onment al eval uat i on Fi ksel and Wapman (1994) Vi vona (1994)
Eval uat i on of t echnol ogy devel opment s Cur r y (1993) Vacek, Hopki ns and M acPher so n
(1995)
Pr oduct i o n/ manuf act ur i ng Wi l ki nson (1990) M al kki , Enwal d and Toi vonen
(1991)
Rel i abi l i t y anal ysi s Zhou and Cai (1994)
Fai l ur e anal ysi s Rooney and Jackson (1996)
Avai l abi l i t y anal ysi s Fai r cl ough (1989)
M ai nt enance Lansdowne (1994) Di nesh Kumar (2000)
M ai nt ai nabi l i t y Govi l (1992), Di nesh Kumar (2000)
Condi t i on moni t or i ng Hut t on (1994)
Logi st i cs suppor t anal ysi s M cAr t hur and Snyder (1989)
Oper at i on and suppor t Cur r y (1989) Snyder (1990) St one, Dr ubka and
Br aun (1994)
Tr anspor t at i on Wonsi ewi cz (1988) Tzemos (1990)
Val ue engi neer i ng Har di ng (1996)
Li f ecycl e cost benef i t anal ysi s Adl er , Her kamp, Wi esl er and Wi l l i ams (1995)
Pl anning and budget i ng
Pr ocur ement st r at egy Pr of i t t (1994)
Busi ness pl anni ng Jones (1994)
Budget ar y pr ovi si on Ki r kpat r i ck (1995)
M anpow er , per sonnel and t r ai ni ng pl anni ng Col e (1991)
Cost management and cont rol
Cost management Fabr ycky and Bl anchar d (1991)
Desi gn f or / t o Cost M i chael s and Woo d (1989) Dean and Unal (1991)
Proj ect / program management and cont rol
Pr ogr am management Gr eene (1991) Zhi (1993)
Pr oj ect cont r ol Gobl e and Paul (1995)
58.1 Table 1 Continued
Management
Lif ecycl e management Hel l (1995)
Physi cal asset management Hodges (1996) Sher wi n (1996)
Act i vi t ybased management Br i mson and Ant os (1994)
Cont r act i ng
269
Cont r act pr ovi si on Aksel sson and Bur st r om (1994) Baat he (1995)
Mar ket i ng
M ar ket i ng of commer ci al pr oduct s Car r uba (1992)
59.
270
Li st of Ref er ences
Adl er , D.J., Her kamp, J.A., Wi esl er , J.R. and Wi l l i ams, S.B. (1995) ' Li f eCycl e Cost and Benef i t s of Pr ocess
Aut omat i on i n Bul k Phar maceut i cal s' , ISA Tr ansact i ons, Vol . 34, No. 2, pp. 133139.
Ahmed, N. U. (1995) ‘ A desi gn and Impl ement at i on M odel f or Li f e Cycl e Cost M anagement Syst em’ ,
Inf or mat i on and M anagement , Vol . 28, pp. 261269.
Aksel sson, H. and Bur st r om, B. (1994) ' Li f eCycl e Co st Pr ocur ement of Sw edi sh St at e Rai l w ays Hi gh
Speed Tr ai n X2000' , Pr oceedi ngs of t he Inst i t ut i on of M echani cal Engi neer s, Par t F, Jour nal of Rai l and
Rapi d Tr ansi t , Vol . 208, No. 1, pp.5159.
Asi edu, Y. and Gu, P. (1998) ‘ Pr oduct Li f e Cycl e Cost Anal ysi s: St at e of t he Ar t Revi ew’ , Int er nat i onal
Jour nal of Pr oduct i on Economi cs, Vol. 36, No. 4, 883908.
Baat he, O. (1995) ‘ Li f e Cycl e M anagement f or t he A/ C JAS 39 Gri ppen’ , Pr oceedi ngs of 11t h Int er nat i onal
Congr ess, Soci et y of Logi st i cs Engi neer s, St ockhol m, pp. 159167.
Bl anchar d, B.S. (1972) ‘ Desi gn and M anage t o Li f e Cycl e Cost ’ , M / A Pr ess, Or egon, USA.
Bl anchar d, B.S. (1986) Logi st i cs Engi neer i ng and M anagement , 3r d Edi t i on, Pr ent i ceHal l , Engl ew ood
Cl i f f s, New Jer sey.
Bl anchar d, B.S. (1998) Syst em Engi neer i ng M anagement , 2nd Edi t i on, John Wi l ey, New Yor k.
Bl anchar d, B.S. and Fabr ycky, W.J. (1990) Syst ems Engi neer i ng and Anal ysi s, 2nd Edi t i on, Pr ent i ceHal l ,
Engl ewood Cl i f f s, New Jer sey.
Br i mson, J.A. and Ant os, J. (1994) Act i vi t yBased M anagement , John Wi l ey, New Yor k.
Bur ns, J. W. (1994) ‘ Ai r cr af t Cost Est i mat i on M et hodol ogy and Val ue of a Pound Der i vat i on f or
Pr el i mi nar y Desi gn Devel opment Appl i cat i ons, SAWE Paper No. 2228, Long Beach, USA.
Car r ubba, E.R. (1992) ' Int egr at i ng Li f eCycl e Cost and Cost of Owner shi p i n t he Commer ci al Sect or ' ,
Pr oceedi ngs of t he Annual Rel i abi l i t y and M ai nt ai nabi l i t y Symposi um, pp. 101108.
Checkl and, P. (1981) Syst ems Thi nki ng, Syst ems Pr act i ce, John Wi l ey, New Yor k.
Cur r y, E.E. (1989) ' STEP  A Tool f or Est i mat i ng Avi oni cs Li f e Cycl e Cost s' , IEEE Aer ospace and El ect r oni c
Syst ems M agazi ne, Vol . 4, No. 1, pp. 3032.
Cur r y, E.E. (1993) ' FALCCM H: Funct i onal Avi oni cs Li f e Cycl e Cost M ode f or Har dw ar e' , 1993 IEEE
Nat i onal Aer ospace and El ect r oni cs Conf er ence, Dayt on, Ohi o, M ay 2428, pp.950953.
271
Dacko, L.M . and Dar l i ngt on, R.F. (1988) ' Li f e Cycl e Cost Pr ocedur e f or Commer ci al Ai r cr af t Subsyst em' ,
Pr oceedi ngs of t he Annual Rel i abi l i t y and M ai nt ai nabi l i t y Symposi um, 1988, Los Angel es, Januar y 26
28, pp.389394.
Dahl en, P., and Bol msj o, B.S. (1996) ‘ Li f e Cycl e Cost Anal ysi s of Labour Fact or ,’ Int er nat i onal Jour nal of
Pr oduct i on Economi cs, Vol . 4647, pp. 459467.
Dean, E. B. (1995) ‘ Par amet r i c Cost Depl oyment ’ , Pr oceedi ngs of t he Sevent h Symposi um on Qual i t y
Funct i on Depl o yment , M I, USA, pp. 2734.
Degr aeve, Z., and Roodhoof t F. (1999) ‘ Impr o vi ng t he Ef f i ci ency of t he Pur chasi n g Pr ocess Usi ng t he
Tot al Cost of Owner shi p Inf or mat i on: The Case of Heat i ng El ect r odes at Cocker i l l Sambr e S.A.,’
Eur opean Jour nal of Oper at i onal Resear ch, Vol . 112, pp. 4253.
De La M ar e, R.F. (1982) ' M anuf act ur i ng Syst ems Economi cs: The Li f e Cycl e Cost s and Benef i t s of
Indust r i al Asset s' , Hol t , Ri nehar t and Wi nst on, London.
Dew hur st . P., and Boot hr oyd, G. (1988) ‘ Ear l y Cost Est i mat i ng i n Pr oduct Desi gn ,’ Jour nal of
M anuf act ur i ng Syst ems,’ Vol. 7, pp. 183191.
Dhi l l on, B.S. (1989) Li f e Cycl e Cost i ng : Techni ques, M odel s and Appl i cat i ons, Gor don and Br each, New
Yor k.
Di enemann, P.F. (1966) ‘ Est i mat i ng Cost Uncer t ai nt y Usi ng M ont e Car l o Techni ques’ , M emor andum RM 
4854PR, The RAND Cor por at i on, Sant a M oni ca, Cali f or ni a, Januar y.
Dowl at shahi , S. (1992) ‘ Pr oduct Desi gn i n Concur r ent Engi n eer i n g Envi r onment : An Opt i mi zat i on
Appr oach’ , Jour nal of Pr oduct i on Resear ch, Vol . 30, 18031818.
Fabr ycky, W.J. (1994) ‘ M od el i ng and Indi r ect Exper i ment at i on i n Syst em Desi gn Eval uat i on’ , The Jour nal
of NCOSE, Vol . 1, No. 1, pp. 133144.
Fabr ycky, W.J. and Bl anchar d, B.S. (1991) Li f eCycl e Cost and Economi c Anal ysi s, Pr ent i ceHal l ,
Engl ewood Cl i f f s, New Jer sey.
Fabr ycky, W.J. and Thuesen, G.J. (1980) Economi c Deci si on Anal ysi s, 2nd edi t i on, Pr ent i ceHal l ,
Engl ewood Cl i f f s, New Jer sey.
Fai r cl ough, M .R. (1989) ' Ar ena  A Sof t w ar e Ai d f or M easur i ng Syst em Avai l abi l i t y and Li f e Cycl e Cost ' ,
Rel i abi l i t y ' 89, Par t 2, London, June 1416, pp.p4B/ 14B/ 1/ 6.
Fi ksel , J. and Wapman, K. (1994) ' How t o Desi gn f or Envi r onment and M i ni mi ze Li f e Cycl e Cost ' ,
Pr oceedi ngs of t he 1994 IEEE Int er nat i onal Symposi um on El ect r oni cs and t he Envi r onment , San
Fr anci sco, pp. 7580.
Fi nkel st ei n, W. and Guer t i n, J.A.R. (1988) Int egr at ed Logi st i cs Suppor t : The Desi gn Engi neer i ng Li nk, IFS
Publ i cat i ons, Kempst on, Bedf or d.
272
Fi sher , G.H. (1971) Cost Consi der at i ons i n Syst ems Anal ysi s, Amer i can El sevi er , New Yor k.
Fl anagan, R. and Nor man, G. (1983), Li f e Cycl e Cost i ng f or Const r uct i on, Quant i t y Sur veyor s Di vi si on of
t he Royal Inst i t ut i on of Char t ed Sur veyor s, London.
Gi bbs, D.J.L. and Ki ng, R.J. (1989) ' Li f e Cycl e Cost i ng i n t he Desi gn of Naval Equi pment ' , Rel i abi l i t y ' 89,
Par t 1, No. 1989, June 1416, pp. 2A/ 5/ 1  2A/ 5/ 8.
Govi l , K.K. (1992) ' A Si mpl e M odel f or Li f eCycl e vs M ai nt ai nabi l i t y Funct i on' , M i cr oel ect r oni cs and
Rel i abi l i t y, Vol . 32, No. 12, pp. 269270.
Gr eene, L. E., and Shaw , B. L. 1990 ‘ The St eps f or Successf ul Li f e Cycl e Cost Anal ysi s,’ Pr oceedi ngs of t he
IEEE 1990 Nat i onal Aer ospace and El ect r oni cs Conf er ence, Dayt on, Ohi o, pp. 12091216.
Gr eves, Y. P., and Schr ei ber , B. (1993) ‘ Engi neer i ng Cost i ng Techni ques i n ESA, Int er net si t e:
ht t p:/ / w w w .esapub.esr i u. esa.i t .poi nt t obul l et / gr eves.ht ml .
Gupt a, Y.P. (1983) ‘ Li f e Cycl e Cost M odel s and Associ at ed Uncer t ai nt i es’ i n Skw i r zynski, J. K. (1983) (Ed.),
El ect r oni c Syst ems Ef f ect i veness and Li f e Cycl e Cost i ng, Spr i nger Ver l ag, Ber l i n, pp. 535  549.
Gupt a, Y., and Chow , W. S. (1985) ‘ Tw ent yf i ve Year s of Li f e Cycl e Cost i ng – Theor y and Appl i cat i on: A
Sur vey’ , The Int er nat i onal Jour nal of Qual i t y and Rel i abi l i t y M anagement , Vol . 2, pp. 551576.
Habayeb, A.R. (1987) Syst em Ef f ect i veness, Per gamon Pr ess, Oxf or d.
Hal l , A.D. (1962) A M et hodol ogy f or Syst ems Engi neer i ng, Van Nost r and, Pr i ncet on, New Jer sey.
Hanssmann, F. (1968) Oper at i ons Resear ch Techni ques f or Capi t al Invest ment , John Wi l ey, New Yor k.
Har t , G.W. (1985) ' Int er pr et at i on of Li f e Cycl e Cost s' , Rel i abi l i t y ' 85  Pr oceedi ngs , Vol . 1, Jul y 1012, pp.
3A. 4. 13A. 4. 5.
Har t , G.W. (1985) ‘ Int er pr et at i on of Li f e Cycl e Cost s’ , Rel i abi l i t y ’ 85  Pr oceedi ngs, Nat i onal Cent r e of
Syst ems Rel i abi l i t y, UKAEA and Inst i t ut e of Qual i t y Assur ance, Vol . 1, 1012 Jul y, pp. 3A.4.1  3A.4.5.
Har vey, G. (1976) ‘ Li f e Cycl e Cost i ng: A Revi ew of t he Techni que’ , M anagement Account i ng, pp. 343347.
Hodges, N.W. (1996) ' The Economi c M anagement of Physi cal Asset s' , M echani cal Engi neer i ng
Publ i cat i ons, London.
Howar d, R.J. (1991) ' Road Li f e Cycl e Cost i ng' , Nat i onal Conf er ence Publ i cat i on  Inst i t ut i on of Engi neer s,
Aust r al i a, No. 91, Par t 14, pp. 5559.
Jayakr i shnan, V. E. (2002) ‘ M odel f or Fi ght er Ai r cr af t Li f e Cycl e Cost Est i mat i on,’ M .Tech. Thesi s, Indi an
Inst i t ut e of Technol ogy Bombay, Indi a
Johnson, V.S. (1990) ' M i ni mi zi ng Li f eCycl e Cost f or Subsoni c Commer ci al Ai r cr af t ' , Jour nal of Ai r cr af t ,
Vol . 27, No. 2, pp. 139145.
273
Jones, C. (1994) ' Li f e Cycl e Cost M odel s' , Col l oqui um on Li f e Cycl e Cost i ng and t he Busi ness Pl an, London,
Febr uar y 1, pp.2/ 1  2/ 6.
Jor dan, N. (1960) ' Some Thi nki ng About ' Syst em' ' , i n Opt ner (1973), pp. 5372.
Kauf man, R. J. (1970) ‘ Li f e Cycl e Cost i ng: A Deci si on M aki ng Tool of Capi t al Equi pment Acqui si t i on,’ Cost
and M anagement , pp. 2128.
Keene, S.J. and Keene, K.C. (1993) ' Reduci ng t he Li f e Cycl e Cost of Sof t war e t hr ough Concur r ent
Engi neer i ng' , Pr oceedi ngs of t he 1993 Annual Rel i abi l i t y and M ai nt ai nabi l i t y Symposi um, At l ant a,
Geor gi a, Januar y 2628, pp.305309.
Ki r kpar t i ck, P. S., and Jones, A. L. (1974) ‘ Li f e Cycl e Cost Compar i sons of Avi oni c Syst em Desi gn
Al t er nat i ves,’ Pr oceedi ngs of t he IEEE 1974 Nat i onal Aer ospace and El ect r oni cs Conf er ence, Dayt on,
Ohi o, pp. 514520.
Knezevi c, J. (1993) Rel i abi l i t y, M ai nt ai nabi l i t y and Suppor t abi l i t y : A Pr obabi l i st i c Appr oach, M cGr aw Hi l l
, London.
Knezevi c, J. (1997) Syst ems M ai nt ai nabi l i t y : Anal ysi s, Engi neer i ng and M anagement , Chapman and Hal l,
London.
Knot t s, R. M . H. 1996 ‘ Anal ysi s of t he Impact of Rel i abi l i t y, M ai nt ai nabi l i t y and Suppor t abi l i t y on t he
Busi ness and Economi cs of Ci vi l Ai r Tr anspor t Ai r cr af t M ai nt enance and Suppor t , M . Phi l . Thesi s,
Uni ver si t y of Exet er , UK
Lamar , W.E. (1981) ‘ A Revi ew and Assessment of Syst em Cost Reduct i on Act i vi t i es’ , AGARD Conf er ence
Pr oceedi ngs, No. 289, Desi gn t o Cost and Li f e Cycl e Cost ’ i n Dhi l l on, B.S. (1989) Li f e Cycl e Cost i ng :
Techni ques, M odel s and Appl i cat i ons, Gor don and Br each, New Yor k.
Lamb, R.G. (1995) Avai l abi l i t y Engi neer i ng and M anagement f or M anuf act ur i ng Pl ant Per f or mance,
Pr ent i ceHal l , Engl ewood Cl i f f s, New Jer sey.
Lansdowne, Z.F. , (1994) ' Bui l t InTest Fact or s i n a Li f eCycl e Cost M odel ' , Rel i abi l i t y Engi neer i ng and
Syst em Saf et y, Vol . 43, No. 3, pp. 325330.
M acedo, M .C., Dobr ow , P. V., and O’ Rour ke, J J. (1978) ‘ Val ue M anagement f or Const r uct i on,’ John
Wi l ey & Sons, Chi est er , UK.
M al kki , H., Enwal d, P. and Toi vonen, J. (1991) ' Exper i ence of Tr ansf er r i ng Li f e Cycl e Cost i ng t o
M anuf act ur i ng Indust r y' , Amer i can Soci et y of M echani cal Engi neer s, pp. 18.
M cNi chol s, G. R. (1983a) A Repor t ed Pr esent at i on on Cost Gr ow t h i n Skw i r zynski , J. K. (1983) (Ed.),
El ect r oni c Syst ems Ef f ect i veness and Li f e Cycl e Cost i ng, Spr i nger Ver l ag, Ber l i n, pp. 567  582.
M cNi chol s, G. R. (1983b) ‘ Uncer t ai nt i es of LCC Pr edi ct i ons’ i n Skw i r zynski , J. K. (1983) (Ed.), El ect r oni c
Syst ems Ef f ect i veness and Li f e Cycl e Cost i ng, Spr i nger Ver l ag, Ber l i n, pp. 583  594.
274
M i chael s, J.V. and Wood, W.P. (1989) Desi gn t o Cost , John Wi l ey, New Yor k.
Ni chol as, C.S. (1999) ‘ Li f e Cycl e Cost i ng’ , M .Phi l . Thesi s, Uni ver si t y of Exet er , UK
Ni chol as, C.S. (1994) ‘ Det er mi ni ng and Achi evi ng Oper at i onal Avai l abi l i t y  Economi c Impact ’ ,
Pr oceedi ngs of 10t h Int er nat i onal Congr ess, Soci et y of Logi st i cs Engi n eer s, Exet er , M ar ch, pp. 275
283.
Ni chol as, C.S. (1996a) ‘ The Impact of Di scount i ng Li f eCycl e Cost on Suppor t abi l i t y Engi neer i ng’ ,
Communi cat i ons i n Rel i abi l i t y, M ai nt ai nabi l i t y, Suppor t abi l i t y and Logi st i cs’ , Vol . 3, No. 1, pp. 4246.
Ni chol as, C.S. (1996b) ‘ Suppor t abi l i t y Engi neer i ng  The Economi c Di mensi on’ , Pr oceedi ngs of t he Soci et y
of Logi st i cs Engi neer s Annual Int er nat i onal Symposi um, Washi ngt on, pp. 151157.
Nobl e, J. S., and Tanchoco, J. M . A. (1990) ‘ Concur r ent Desi gn and Economi c Just i f i cat i on i n Devel opi ng a
Pr oduct ’ , Int er nat i onal Jour nal of Pr oduct i on Resear ch, Vol . 28, pp. 12251238.
Ost w al d, P.F. (1992) Cost Est i mat i ng, 3r d edi t i on, Pr ent i ceHal l , Engl ewood Cl i f f s, New Jer sey.
Pl ebani , S. , Rosi , R. and Zanet t a, L. (1996) ' Li f e Cycl e Cost s Compar i son and Sensi t i vi t y Anal ysi s f or
M ul t i medi a Net wor ks' , Pr oceedi ngs of t he 1996 IEE Col l oqui um on Opt i cal and Hybr i d Access
Net w or ks, No. 052, London, M ar ch 45, pp. 3/ 13/ 6.
Pr of f i t t , J.T. (1994) ' Li f e Cycl e Cost of Ai r cr af t Syst ems' , Col l oqui um on Li f e Cycl e Cost i ng and t he
Busi ness Pl an, IEE Col l oqui um (Di gest ), No. 013, pp. 1/ 11/ 3.
Per er a, H.S.C., Nagar ur , N., and Tabucanon, M .T. (1999) ‘ Component Par t St andar di zat i on: A Way t o
Reduce t he Li f e Cycl e Cost s of Pr odu ct s,’ Int er nat i onal Jour nal of Pr oduct i on Economi cs, ’ Vol . 6061,
pp. 109116.
Quade, E.S. (1963b) ‘ Th e Li mi t at i ons of A Cost Ef f ect i veness Appr oach t o M i l i t ar y Deci si onM aki ng’ ,
Repor t P2798, The RAND Cor por at i on, Sant a M oni ca, Cal i f or ni a.
Quade, E.S. (1965) ‘ Cost Ef f ect i veness Anal ysi s: An Appr eci at i on, Repor t P3248, The RAND Cor por at i on,
Sant a M oni ca, Cal i f or ni a.
Raymer , D.P. (1992) Ai r cr af t Desi gn: A Concept ual Appr oach, Amer i can Inst i t ut e of Aer onaut i cs and
Ast r onaut i cs, Washi ngt on.
Ri ggs, J.L. and Jones, D. ‘ Fl owgr aph r epr esent at i on of Li f e Cycl e Cost M et hodol ogy  A New Per spect i ve
f or Pr oj ect M anager s’ , IEEE Tr ansact i ons on Engi neer i ng M anagement , Vol . 37, No.2, pp.147152.
Ri ggs, J.L. (1968) Economi c Deci si on M odel s f or Engi neer s and M anager s, M cGr aw Hai l l , New Yor k.
Rooney, C. and Jackson, E. (1996) ' Uni t Fai l ur e: A Li f e Cycl e Cost Anal ysi s' , Gl ass Di gest , Vol . 75, No.2,
pp.7.
275
Rose, A. (1984) ' Li f e Cycl e Cost i ng Today' , 28t h Annual M eet i ng of t he Amer i can Associ at i on of Cost
Engi neer s and t he Ei ght h Int er nat i onal Cost Engi neer i ng Congr ess, M ont r eal , June 2427, pp. J.3.1  J.
3. 4.
Roskam, J. (1990) ‘ Ai r pl ane Desi gn Par t VII I: Air pl ane Cost Est i mat i on Desi gn, Devel opment ,
M anuf act ur i ng and Oper at i ng, Roskam Avi at i on and Engi neer i ng Cor por at i on Repor t , Kansas, USA.
Sabbagh, K. (1995) 21st Cent ur y Jet : The M aki ng of t he Boei ng 777, M acmi l l an, London.
SAE (1992) ‘ Li f e Cycl e Cost  Techni ques and Appl i cat i ons’ , Aer ospace Recommended Pr act i ce, SAE
ARP4293, SAE, War r endal e.
Scher er , F.M . (1964) The Weapons Acqui si t i on Pr ocess : Economi c Incent i ves, Har var d Uni ver si t y,
Har var d.
Sher i f , Y.S. and Kol ar i k, W. J. (1981) ‘ Li f e Cycl e Cost i ng: Concept and Pr act i ce’ , OM EGA, The Int er nat i onal
Jour nal of M anagement Sci ence, Vol . 9, No. 3, pp. 287  296.
Sher wi n, D. (1996) ‘ Economi c and Account i ng Asp ect s Of Ter ot echnol ogy and Li f eCycl e Cost i ng’ ,
Pr oceedi ngs of t he Int er nat i onal Conf er ence of M ai nt enance Soci et i es, M el bor ne, M ay 2223, Paper
46.
Shi el ds, M . D., and Young, S. M . 1991 ‘ M anagi ng Pr oduct Li f e Cycl e Cost : An Or gani zat i onal M odel , Cost
M anagement , pp. 3952.
Si mon, H. A. (1959) ‘ Theor i es of Deci si on M aki ng i n Economi cs and Behavi our al Sci ence’ , Amer i can
Economi c Revi ew, June.
Si ngl et on, W.T. (1974) M anM achi ne Syst ems, Pengui n, Har mondswor t h.
Snyder , H.M . (1990) ' Li f e Cycl e Cost M odel f or Dor mant Syst ems' , Pr oceedi ngs of t he IEEE 1990 Nat i onal
Aer ospace and El ect r oni cs Conf er ence  NAECON 1990, Vol . 3, Dayt on, M ay 2125, pp.12171219.
Sr i ni vasan, N. K., Nagesw ar a A. M ., and Pr abhakar B K. (1997) ‘ CASSA M odel f or Li f e Cycl e Cost
Est i mat i on f or Ai r cr af t ,’ Repor t No. S971, CASSA, DRDO, Bangal or e.
St ahl , L. and Wal l ace, M . (1995) ' Li f e Cycl e Cost Compar i son  Tr adi t i onal Cool i ng Syst ems vs. Nat ur al ' ,
Pr oceedi ngs of t he 1995 17t h Int er nat i onal Tel ecommuni cat i ons Ener gy Conf er ence, Hague, Oct ober
29  November 1, pp.259265.
St one, K.W., Dr ubka, R.E.and Br aun, H. (1994) ' Impact of St i r l i ng Engi ne Oper at i onal Requi r ement s on
Di sh St i r l i ng Syst em Li f e Cycl e Cost s' , Pr oceedi ngs of t he 1994 ASM E/ JSM E/ JSES Int er nat i onal Sol ar
Ener gy Conf er ence, San Fr anci sco, M ar ch 2730, pp.529534.
St ump, E. J. (1988) ‘ An Appl i cat i on of M ar kov Chai ns t o Li f e Cycl e Cost Anal ysi s,’ Engi neer i ng Cost s and
Pr oduct i on Economi cs, Vol. 14, pp. 151156.
276
Takagi shi , S.K. (1989) ' El ect r i c Power ver sus Pet r ol , M et hanol or Gas. Li f e Cycl e Cost Compar i son' , 9t h
Int er nat i onal El ect r i c Vehi cl e Symposi um, El ect r i c Vehi cl e Devel opment s,Vol . 8, No. 3, Tor ont o,
November 1315, pp.7781.
Tayl or , W.B. (1981) ‘ The Use of Li f e Cycl e Cost i ng i n Acqui r i ng Physi cal Asset s’ , Long Range Pl anni ng, Vol .
14, pp. 3243.
Thuesen, G.J. and Fabr ycky, W.J. (1989) Engi neer i ng Economy, 7t h edi t i on, Pr ent i ceHal l , Engl ew ood
Cl i f f s, New Jer sey.
UK Depar t ment of Indust r y (1977) ‘ Li f e Cycl e Cost i n g i n t he M anagement of Asset s' , Depar t ment of
Indust r y, Commi t t ee f or Ter ot echnol ogy, H.M .S.O., London.
US Depar t ment of Def ense, (1983a) ‘ Desi gn t o Cost ’ , Di r ect i ve 4245.3, US Depar t ment of Def ense.
US Depar t ment of Def ense, (1983b) ‘ Li f e Cycl e Cost i n Navy Acqui si t i ons’ , M ILHDBK259 (Navy), US
Depar t ment of Def ense.
US Depar t ment of Def ense, (1996) ‘ M aj or Syst ems Acqui si t i ons’ , Di r ect i ve 5000.1 (Revi sed), US
Depar t ment of Def ense.
Vacek, R.M ., Hopki ns, M . and M acPher son, W.H. (1996) ' Devel op ment , Demonst r at i on and Int egr at i on
of Advanced Technol ogi es t o Impr ove t he Li f e Cycl e Cost s of Space Syst ems' , Pr oceedi ngs of t he 1996
IEEE Aer ospace Appl i cat i ons Conf er ence, Vol . 1, IEEE, Los Al amos, pp. 217225.
Ver ma, D. (1995) ‘ Con cept ual Syst em Desi gn Synt hesi s, Anal ysi s and Eval uat i on’ , Techni cal M onogr aph
1, Cent r e f or M anagement of Indust r i al Rel i abi l i t y, Cost and Ef f ect i veness, Facul t y of Engi neer i ng,
Uni ver si t y of Exet er , Exet er .
Vi vona, M .A. (1994) ‘ Audi t Envi r onment al Pr ocesses Usi ng Li f eCycl e Cost s’ , Hydr ocar bon Pr ocessi ng,
Vol . 73, No. 8, p. 115.
Whi t e, G. E., and Ost w al d, P. H. (1976) ‘ M anagement Account i ng’ , pp. 3942.
Whi t e, J. A., Agee, M .H. and Case, K.E. (1977) Pr i nci pl es of Engi neer i ng Economi c Anal ysi s, John Wi l ey,
New Yor k.
Wi l ki nson, V.K. (1990) ' Li f e Cycl e Cost Anal yses of Gover nment Pr oduct i on and Ser vi ces' , Pr oceedi ngs of
t he 1990 Int er nat i onal Indust r i al Engi neer i ng Conf er ence, No. 1990, San Fr anci sco, M ay 2023, pp.
111116.
Wi nkel , J. D. (1996) ' Use of Li f e Cycl e Cost i ng i n New and M at ur e Appl i cat i ons' , Pr oceedi ngs of t he 1996
NPF/ SPE Eur opean Pr oduct i on Oper at i ons Conf er ence, Soci et y of Pet r ol eum Engi neer s (SPE),
St avanger , Nor w ay, Apr i l 1617, pp. 239248.
Wonsi ewi cz, T.J. (1988) ' Li f e Cycl e Cost Anal ysi s. Key Assumpt i ons and Sensi t i vi t y of Resul t s' ,
Tr anspor t at i on Resear ch Recor d, Lane Ent er pr i ses Inc, Camp Hi l l , pp.113117.
Woodw ar d D. G. (1997) ‘ Li f e Cycl e Cost i ng – Theor y, Inf or mat i on, Acqui si t i on and Appl i cat i on,’
Int er nat i onal Jour nal of Pr oj ect M anagement , Vol . 15, No. 6, pp. 335344.
277
Woud, J.K., Smi t , K. and Vuci ni c, B. (1997) ' M ai nt enance Pr ogr amme Desi gn f or M i ni mal Li f e Cycl e Cost s
and Accept abl e Saf et y Ri sks' , Int er nat i onal Shi pbui l di ng Pr ogr ess, Vol . 44, No. 437, Del f t Uni ver si t y
Pr ess, Del f t , Net her l ands, pp.77100.
Zhi , H. (1993) ' Si mul at i on Anal ysi s i n Pr oj ect Li f e Cycl e Cost ' , Cost Engi neer i ng, Vol . 35, No. 12,
M or gant own, West Vi r gi ni a, pp. 1317.
Zhou, X. and Cai , N. (1994) ' Dynami c St udy of t he Rel at i onshi p bet ween Li f e Cycl e Cost and Rel i abi l i t y i n
El ect r oni c Pr oduct ' , Ti en Tzu Hsueh Pao/ Act a El ect r oni ca Si ni ca, Vol . 22, No. 11, Hangzhou Inst . of
El ect r oni cs Engi neer i ng, Hangzhou, Chi na, pp.2227.
278
Tot al Cost of Ownershi p f or Rai l way Asset s: A Case St udy on BOXN Wagons of Indi an Rai l ways
U Di nesh Kumar * , Gopi nat h Chat t opadhyay* * and H S Pannu* * *
* Indi an Inst i t ut e of M anagement Cal cut t a, Joka, D H Road, Kol kat a 700104, INDIA
* * Dept . of M echani cal , M anuf act ur i ng and M edi cal Engi neer i ng, Queensl and Uni ver si t y of Technol ogy,
Br i sbane, QLD 4001, Aust r al i a
* * * Sout h East er n Rai l w ay, Gar den Reach Road, Kol kat a 700043, INDIA
di nesh@i i mcal .ac.i n, g.chat t opadhyay@qut .edu.au, h_s_pannu @yahoo.co.i n
ABSTRACT
Deci si on t o pur chase any capi t al equi pment must be based on i t s t ot al cost of ow ner shi p (TCO) r at her
t han t he usual pr act i ce of pr ocur ement based on t he i ni t i al pur chase pr i ce. In t he r ecent year s, TCO has
become a par t of t he st r at egi c cost management and t he concept can be appl i ed f or ef f ect i ve
pr ocur ement of r ai l w ay asset s. TCO pr ovi des an i nsi ght i nt o t he t ot al cost of acqui si t i on and sust enance
and t hus ef f ect i vel y suppor t deci si onmaki ng i n eval uat i on of var i ous al t er nat i ves. The pr i mar y
obj ect i ve of t hi s pap er i s t o devel op model s f or pr edi ct i on of cost of ow ner shi p of capi t al asset s. The
model s ar e devel oped usi ng t he M ar ko v and r enewal pr ocesses dependi n g on t he t i me t o f ai l ur e
di st r i but i on of i ndi vi dual i t ems w i t hi n t he capi t al equi pment . The model s devel oped i n t hi s paper ar e
val i dat ed usi ng t he dat a f r om BOXN wagon used by t he Indi an Rai l ways.
Keywor ds: Asset management , M ai nt enance, M ar kov and r enewal pr ocesses, Tot al cost of owner shi p.
1. INTRODUCTION
“ Val ue f or M oney” has become one of t he i mpor t ant cr i t er i a i n an i ncr easi ngl y compet i t i ve busi ness
envi r onment . Li f e Cycl e Cost (LCC) and t he Tot al Cost of Ow ner shi p (TCO) ar e t w o i mpor t ant f i nanci al
measur es t hat ar e used f or deci si on maki ng i n acqui si t i ons t o eval uat e t he val u e of any capi t al
equi pment (Hampt on, 2004; Humphr i es, 2004). Li f e cycl e cost r ef er s t o al l cost s associ at ed w i t h t he
279
pr oduct or syst em as appl i ed t o def i ned l i f e cycl e. That i s, st ar t i ng f r om r equi r ement anal ysi s, desi gn,
pr oduct i on, oper at i on and mai nt enance t i l l di sposal . Tot al cost of ow ner shi p (TCO) i s a phi l osophy,
w hi ch i s ai med at under st andi ng t he t r ue cost of buyi ng a par t i cul ar pr oduct or ser vi ce f r om a par t i cul ar
suppl i er . Fr om i t s or i gi ns i n def ence equi pment pr ocur ement i n t he US i n ear l y 1960s, t he use of l i f e
cycl e cost and cost of ow ner shi p has ext ended t o ot her ar eas of t he publ i c and pr i vat e sect or s. LCC and
TCO ar e bei ng used t o assi st i n deci si onmaki ng, budget pl anni ng, cost cont r ol , and r ange of ot her
act i vi t i es t hat occur over t he l i f e of compl ex t echnol ogi cal equi pment .
LCC anal ysi s i s appl i ed r out i nel y t o mi l i t ar y pr oj ect s (Bl anchar d, 1986). I n t he mi l i t ar y sect or t he
consumer , by f undi ng t he pr oj ect and oper at i ng t he r el at ed pr oduct , essent i al l y bear s t he t ot al l i f e cycl e
cost cover i ng t he maj or cost el ement s i n al l st ages of a pr oduct ’ s l i f e cycl e. The t er m LCC anal ysi s i s
r ar el y used i n t he commer ci al sect or . Inst ead, t he mai n f ocus i s on TCO w her e r el at ed cost s, cover i ng
acqui si t i on (pur chase or l ease), oper at i on, mai nt enance and suppor t ar e bor ne by t he cust omer . In
addi t i on, t he cust omer can al so i ncur cost s w hen t he pr oduct i s not avai l abl e f or use, t hat i s, ‘ down t i me
cost s’ .
The obj ect i ves of LCC/ TCO ar e (Fl anagan and Nor man, 1983):
• To enabl e i nvest ment opt i ons t o be mor e ef f ect i vel y eval uat ed.
• To consi der t he i mpact of all cost s r at her t han onl y t he i ni t i al capi t al cost s.
• To assi st i n t he ef f ect i ve management of compl et ed pr oj ect s.
• To f aci l i t at e choi ce bet w een compet i ng al t er nat i ves.
In t he Def ence i ndust r y t he syst em’ s l i f e cycl e i s di vi ded i nt o var i ous phases, w hi ch al l ow pr oper
pl anni ng and cont r ol of a pr oj ect . The number of phases d epends on t he nat ur e of t he pr oj ect , pur pose
and whet her t hey ar e appl i ed t o commer ci al , mi l i t ar y or space pr oj ect s (Knot t s, 1998). Commonl y used
phases ar e:
8. Requir ement s (Funct i onal Speci f i cat i on).
9. Concept / Feasi bi l i t y St udi es.
10. Desi gn and Devel opment .
11. Pr oduct i on.
12. Test i ng and Cer t i f i cat i on.
13. Oper at i on, M ai nt enance and Suppor t .
14. Di sposal
It i s r epor t ed by t he US Depar t ment of Def ence t hat 70% of w eapon syst em l i f e cycl e cost i s
commi t t ed by t he end of concept st udi es, 85% by t he end of syst em def i ni t i on and 95% by t he end of
f ul l scal e devel opment (Knot t s, 1998). The US Depar t ment of Def ence has f or mal l y used t he concept of
l i f e cycl e cost i n w eapon syst em acqui si t i on si nce t he ear l y 1960s t hr ough l i f e cycl e cost i ng and l i f e cycl e
cost anal ysi s.
The cost of ow ner shi p appr oach i d ent i f i es al l f ut ur e cost s and r educes t hem t o t hei r pr esent
val ue by use of t he di scoun t i ng t echni ques t hr ough w hi ch t he economi c w or t h of a pr oduct or pr oduct
280
opt i ons can be assessed. In or der t o achi eve t hese obj ect i ves t he f ol l ow i ng el ement s of cost of
owner shi p have been i dent i f i ed (Woor war d, 1997):
• Ini t i al capi t al cost s
• Li f e of t he asset
• The di scount r at e
• Oper at i ng and mai nt enance cost s
• Di sposal cost
1.2 Ini t i al capi t al cost s
The i ni t i al capi t al cost s can be di vi ded i nt o t hr ee subcat egor i es of cost namel y: (1) pur chase cost s,
(2) acqui si t i on/ f i nance cost s, and (3) i nst al l at i on/ commi ssi oni ng/ t r ai ni ng cost s. Pur chase cost s w i l l
i ncl ude assessment of i t ems such as l and, bui l di ngs, f ees, and equi pment . Fi nance cost s i ncl ude
al t er nat i ve sour ces of f unds. Basi cal l y, t he i ni t i al capi t al cost cat egor y i ncl udes al l t he cost s of buyi ng
t he physi cal asset and br i ngi ng i t i nt o oper at i on.
1. Li f e of t he Asset
The est i mat ed l i f e of an asset has a maj or i nf l uence o n l i f e cycl e cost anal ysi s. Fer r y et al (1991) has
def i ned t he f ol l ow i ng f i ve possi bl e det er mi nant s of an asset ’ s l i f e expect ancy:
• Funct i onal l i f e – t he per i od over w hi ch t he need f or t he asset i s ant i ci pat ed.
• Physi cal l i f e – t he per i od over w hi ch t he asset may be expect ed t o l ast physi cal l y, t o w hen
r epl acement or maj or r ehabi l i t at i on i s physi cal l y r equi r ed.
• Technol ogi cal l i f e – t he per i od unt i l t echni cal obsol escence di ct at es r epl acement due t o t he
devel op ment of a t echnol ogi cal l y super i or al t er nat i ve.
• Economi c l i f e – t he per i od unt i l economi c obsol escence di ct at es r epl acement w i t h a l ow er
cost al t er nat i ve.
• Soci al and l egal l i f e – t he per i od unt i l human desi r e or l egal r equi r ement di ct at es
r epl acement .
1.4 The di scount rat e
As t he cost of ow ner shi p i s di scount ed t o t hei r pr esent val ue, sel ect i on of a sui t abl e di scount r at e i s
cr uci al f or TCO anal ysi s. A hi gh di scount r at e w i l l t end t o f avour opt i ons w i t h l ow capi t al cost , shor t l i f e
and hi gh r ecur r i ng cost , whi l st a l ow di scount r at e wi l l have t he opposi t e ef f ect .
281
1.5 Oper at i ons and Mai nt enance Cost s
Cost of ow ner shi p, i n many cases, i s about oper at i on and mai nt enance cost . Est i mat i on of
oper at i on and mai nt enance cost s i s t he essent i al t o mi ni mi se t he t ot al cost of ow ner shi p of t he asset . I n
t he whol e of TCO anal ysi s, est i mat i on of oper at i on and mai nt enance i s t he most chal l engi ng t ask.
1.6 Di sposal cost
Thi s i s t he cost i ncur r ed at t he end of as asset ’ s w or ki ng l i f e i n di sposi ng of t he asset . The
di sposal cost woul d i ncl ude t he cost of demol i t i on, scr appi ng or sel l i ng t he asset .
1.7 Uncer t ai nt i es and Sensi t i vi t y Anal ysi s
LCC/ TCO i s hi ghl y dependent on t he assumpt i ons and est i mat es made whi l e col l ect i ng dat a.
Even t hough i t i s possi bl e t o i mpr ove t he qual i t y of t hese est i mat es, t her e i s al ways an el ement of
uncer t ai nt y associ at ed wi t h t hese est i mat es and assumpt i ons. M acedo et al (1978) i dent i f i es t he
f ol l ow i ng f i ve maj or sour ces of uncer t ai nt y:
6. Di f f er ences b et w een t he act ual and expect ed per f or mance of t he syst em coul d af f ect f ut ur e
oper at i on and mai nt enance cost .
7. Changes i n oper at i onal assumpt i ons ari si ng f r om modif i cat i ons i n user act i vi t y.
8. Fut ur e t echnol ogi cal advances t hat coul d pr ovi d e l ow er cost al t er nat i ves and hence shor t en t he
economi c l i f e of any syst em/ subsyst em.
9. Changes i n t he pr i ce l evel s of maj or r esour ces such as ener gy or manpower , r el at i ve t o ot her
r esour ces can af f ect f ut ur e al t er at i on cost s.
10. Er r or i n est i mat i ng r el at i onshi ps, pr i ce r at es f or speci f i c r esour ces and t he r at e of i nf l at i on i n
over al l cost s f r om t he t i me of est i mat i on t o t he avai l abi l i t y of t he asset .
Whi l e under t aki ng a LCC/ TCO anal ysi s, t her e may be some key par amet er s about whi ch uncer t ai nt y
exi st s, usual l y because of t he i nadequacy of t he i npu t dat a. Bl anchar d (1972) suggest ed t he f ol l owi ng
shoul d be t he subj ect of sensi t i vi t y anal ysi s:
1. Fr equency of t he mai nt enance f act or .
2. Var i at i on of t he asset ’ s ut i l i zat i on or oper at i ng t i me.
3. Ext ent of t he syst em’ s sel f di agnost i c capabi l i t y.
4. Var i at i on of cor r ect i ve mai nt enance hour s per oper at i ng hour .
5. Pr oduct demand r at e.
6. The di scount r at e
282
In t hi s paper , w e devel op mat hemat i cal model s f or pr edi ct i on of t ot al cost of ow ner shi p. The r est of
t he paper i s or gani zed as f ol l ows. In Sect i on 2, w e have d evel oped a f r amew or k f or est i mat i ng t he t ot al
cost of ow ner shi p. Sect i on 3 deal s w i t h t he mat hemat i cal model s f or t ot al cost of ow ner shi p. A case
st udy on BOXN wagon used by t he Indi an Rai l w ays i s used t o i l l ust r at e t he mat hemat i cal model
devel oped i n t he paper .
2. Framework f or Tot al Cost of Ownershi p Model
Tot al cost of ow ner shi p i s dr i ven by r el i abi l i t y, mai nt ai nabi l i t y and suppor t abi l i t y. The obj ect i ve of
t ot al cost of ow ner shi p i s t o mi ni mi se TCO by opt i mi zi ng r el i abi l i t y, mai nt ai nabi l i t y and suppor t abi l i t y.
Fi gur e 1 i l l ust r at es t he r el at i onshi p bet ween t he syst em oper at i onal ef f ect i veness and ot her desi gn
par amet er s (Di nesh Kumar 2000). Tot al cost of ow ner shi p wi l l decr ease as t he r el i abi l i t y i ncr eases.
Si mi l ar l y bet t er mai nt ai nabi l i t y and suppor t abi l i t y woul d decr ease t he mai nt enance and suppor t cost
and hence w i l l decr ease t he t ot al cost of ow ner shi p. How ever , i ncr easi ng r el i abi l i t y, mai nt ai nabi l i t y and
suppor t abi l i t y may r equi r e addi t i onal r esour ces dur i ng t he desi gn and pr oduct devel op ment st age and
hence i s l i kel y t o i ncr ease t he i ni t i al pr ocur ement cost .
Fi gur e 1. Cause and ef f ect dependency bet w een oper at i onal ef f ect i veness, t ot al cost of ow ner shi p and
ot her desi gn par amet er s
The f r amew or k f or cal cul at i ng t ot al cost of ow ner shi p can be ver y compl ex dependi ng on t he
pr ocur ement and asset management st r at egi es used by t he user . In t hi s paper , w e mai nl y f ocus on
Systems Operational Effectiveness: Systems Operational Effectiveness:
A Cause A Cause and and Effect Dependency Effect Dependency
Reliability/
Supportability/ Maintainability/
Design Cause
Operational Effect
Operation
Logistics Maintenance
Time to
Support (TTS)
Time to
Maintain (TTM)
Time to
Failure (TTF)
System Downtime System Uptime
System
Effectiveness
Operation
Maintenance
Logistics
Process
Efficiency
Operational Operational
Effectiveness Effectiveness
Cost as an Independent Variable (CAIV)/TOC
Reliability
Maintainability
Supportability
Availability
Technical
Effectiveness
Inherent
Performance
Functions
Requirements
Priorities
Systems Operational Effectiveness: Systems Operational Effectiveness:
A Cause A Cause and and Effect Dependency Effect Dependency
Reliability/
Supportability/ Maintainability/
Design Cause
Reliability/
Supportability/ Maintainability/
Design Cause
Operational Effect
Operation
Logistics Maintenance
Operational Effect
Operation
Logistics Maintenance
Time to
Support (TTS)
Time to
Maintain (TTM)
Time to
Failure (TTF)
System Downtime System Uptime
System
Effectiveness
Operation
Maintenance
Logistics
Process
Efficiency
System
Effectiveness
Operation
Maintenance
Logistics
Process
Efficiency
Operational Operational
Effectiveness Effectiveness
Cost as an Independent Variable (CAIV)/TOC
Operational Operational
Effectiveness Effectiveness
Cost as an Independent Variable (CAIV)/TOC
Reliability
Maintainability
Supportability
Availability
Technical
Effectiveness
Inherent
Reliability
Maintainability
Supportability
Availability
Technical
Effectiveness
Inherent
Performance
Functions
Requirements
Priorities
Performance
Functions
Requirements
Priorities
283
pr ocur ement , oper at i on, mai nt enance and di sposal cost , w hi ch ar e mor e r el evant f or asset s l i ke
w agons. The f r amew or k show n i n Fi gur e 2 i s used f or eval uat i on of t ot al cost of ow ner shi p.
284
Fi gur e 2: Fr amew or k f or cal cul at i on of t ot al cost of owner shi p
3. Mat hemat i cal Model s f or Est i mat i on of Tot al Cost of Ownershi p
In t hi s sect i on w e devel op mat hemat i cal model s f or est i mat i on of var i ous cost el ement s i n t he
t ot al cost of ow ner shi p. The mai n f ocus i s on est i mat i on of i nser vi ce cost . Si nce al l t he cost el ement s i n
t he t ot al cost of own er shi p need t o be di scount ed t o t hei r pr esent val ue, al l t he cost s model s expl ai ned
i n t he subsequent sect i ons ar e cal cul at ed on annual basi s and f i nal l y di scount ed usi ng appr opr i at e
di scount r at e.
3.1 Est i mat i on of operat i ng cost
The oper at i ng cost can be di vi ded i nt o t w o cat egor i es, di r ect oper at i ng cost and over head cost s.
The di r ect oper at i ng cost i s d et er mi ned by t he r esour ces, w hi ch ar e r equi r ed f or oper at i ng t he asset .
The mai n r esour ces f or most of t he syst em ar e ener gy consumed by t he asset and t h e manpower
r equi r ed t o oper at e t he asset . The ener gy consumed by t he asset wi l l depend on t he oper at i onal
avai l abi l i t y of t he syst em cal cul at ed on annual basi s. The oper at i onal avai l abi l i t y, A
o
, of any asset i s
gi ven by (D Kumar et al 2000):
DT MTBM
MTBM
A
O
+
· (1)
wher e:
Acquisition Cost
Operating Cost
Maintenance Cost Logistics Cost
Support Cost
InService Cost
Total Cost of Ownership
285
M TBM = M ean t i me bet w een mai nt enance
DT = Dow n Ti me
M ean t i me bet ween mai nt enance f or dur at i on, T, i s gi ven by:
sm
T
T
T M
T
MTBM
+
·
) (
(2)
Wher e, M (T), i s t he number of f ai l ur es r esul t i ng i n unschedul ed mai nt enance and T
sm
i s t he t i me
bet ween schedul ed mai nt enance. The down t i me, DT, can be est i mat ed usi ng t he f ol l ow i ng equat i on:
sm
sm
T
T
T M
MPMT
T
T
MCMT T M
DT
+
× + ×
·
) (
) (
(3)
Wher e,
M CM T = M ean cor r ect i ve mai nt enance t i me.
M PM T = M ean pr event i ve mai nt enance t i me.
The number of f ai l ur es r esul t i ng i n unschedul ed mai nt enan ce can be eval uat ed usi ng r enewal
f unct i on and i s gi ven by:
∫
− + ·
T
dx x f x T M T F T M
0
) ( ) ( ) ( ) ( (4)
Wher e, F(T) i s t he cumul at i ve di st r i but i on of t he t i met of ai l ur e r andom var i abl e and f (x) i s t he
cor r espondi ng pr obabi l i t y densi t y f unct i on. Equat i on (4) i s val i d onl y w hen t he f ai l ed uni t s ar e r epl aced
or w hen t he r epai r i s as good as new . How ever , i n case of mi ni mal r epai r and i mper f ect r epai r , one may
have t o use model s based on nonho mogeneous Poi sson pr ocess or modi f i ed r enewal pr ocess. For
mor e det ai l s, t he r eader s may r ef er t o Ross (2000). If we assume t hat t he ener gy cost and manpow er
286
cost per uni t t i me i s C
o u
and t he annual usage of t he asset i s T l i f e uni t s. Then t he annual oper at i ng cost
i s gi ven by:
ou O O
C T A C × × · (5)
Assume t hat ‘ r ’ denot es t he di scount r at e. Then t he pr esent val ue of t he oper at i ng cost f or n
t h
per i od (n
t h
year ), C
O,n
, i s gi ven by:
n
ou O
n O
r
C T A
C
) 1 (
,
+
× ×
· (6)
3.2 Est i mat i on of Mai nt enance Cost
The mai n component s of mai nt enance cost ar e cor r ect i ve mai nt enance cost s, pr event i ve
mai nt enance cost s and over haul cost s. M ai nt enance r esour ces t hat used i n per f or mi ng t hat par t i cul ar
mai nt enan ce dr i ve t hese cost s. The mai nt enance cost , C
M
, can be est i mat ed usi ng t he f ol l ow i ng
equat i on:
∑ × + × + × ·
·
k
i
i OH n i pm
sm
cm M
C C
T
T
C T M C
1
, ,
) ( δ (7)
Wher e:
C
cm
= Aver age cost of cor r ect i ve mai nt enance.
C
pm
= Aver age cost of pr event i ve mai nt enance
¹
'
¹
·
otherwise , 0
n period during out carried is i type of overhaul if , 1
, n i
δ
C
OH,i
r epr esent s t he aver age cost of over haul of t ype i . Thi s cost wi l l be added t o t he mai nt enance cost ,
i f t he t ype i over haul i s car r i ed out dur i ng per i od n.
The mai nt enance cost f or per i od ‘ n’ i s gi ven by:
287
,
_
¸
¸
∑ + × + ×
+
·
·
i OH
k
i
n i pm
sm
cm
n
n M
C C
T
T
C T M
r
C
,
1
, ,
) (
) 1 (
1
δ (8)
3.3 Est i mat i on of Logi st i c Support Cost
Logi st i c suppor t cost cover s t he cost s associ at ed w i t h mai nt ai ni ng spar e par t s, mai nt enance
f aci l i t i es, t est equi pment and ot her l ogi st i cs cost s such as t r anspor t at i on cost s. The spar e par t s
cont r i but e si gni f i cant por t i on of t he t ot al supp or t cost . The number of spar es st ocked al so pl ays a
cr uci al r ol e i n t he oper at i onal avai l abi l i t y of t he syst em. Pr act i t i on er s deci de on t he number of spar e
par t s t o be pur chased based on t he t ar get f i l l r at e, α (pr obabi l i t y t hat a demand f or a par t i cul ar spar e
par t can be achi eved f r om t he avai l abl e st ock). Usual l y t he t ar get f i l l r at e i s 85%. Assume t hat N
s
r epr esent t he mi ni mum number of spar es t hat shoul d be st ocked t o achi eve a t ar get f i l l r at e α. Then
t he val ue of N
s
can be cal cul at ed usi ng t he f ol l owi ng equat i on:
∑
× −
·
s
N
k
k
k
T T
0
!
) ( ) exp( λ λ
(9)
The above equat i on i s val i d onl y wh en t he t i met of ai l ur e di st r i but i on f ol l ows exponent i al
di st r i but i on, w her e λ i s t he f ai l ur e r at e. When t he t i met of ai l ur e di st r i but i on i s ot her t han exponent i al ,
t hen w e need t o use r enew al f unct i on t o f i nd t he val ue of N
s
t o achi eve t he t ar get avai l abi l i t y.
The annual l ogi st i cs cost f or t he per i od n, C
L
,n i s gi ven by:
s s
n
n L
C N
r
C ×
+
·
) 1 (
1
,
(10)
3.4 Tot al Cost of Owner shi p
The t ot al cost of ow ner shi p i s obt ai ned by addi ng t he component s gi ven by equat i ons (6), (8)
and (10) over t he desi gned l i f e of t he asset . If t he desi gned l i f e of t he asset i s D, t hen t he t ot al cost of
ow ner shi p, TCO
D
, i s gi ven by:
MF n L
D
n
n M n O P D
C C C C C TCO + + ∑ + + ·
·
] [
,
1
, ,
(11)
288
Wher e C
P
i s t he pr ocur ement pr i ce of w agon and C
M F
i s t he onet i me expenses of mai nt enance
and suppor t equi pment .
4. Case St udy on BOXN wagons used by t he Indi an Rai l ways
Indi an r ai l w ays ar e t he pr i nci pal mode of t r anspor t f or r aw mat er i al f or st eel pl ant s, f i ni shed st eel
f r om st eel pl ant s, coal , i r on, oi l , cement , pet r ol eum pr oduct s, f er t i l i zer and f ood gr ai ns i n I ndi a. Indian
r ai l w ays ow ns mor e t han 5,00,000 w agons. The w agons have a desi gn l i f e of 35 year s. The r equi r ement
of w agons f or f ut ur e i s assessed on t he basi s of f r ei ght t r af f i c pr oj ect ed and t he ant i ci pat ed l evel of
pr oduct i vi t y of wagons measur ed i n t er ms of net t on ki l omet er s (NTKM ) per wagon per day l i kel y t o be
achi eved. The w agons ar e pr ocur ed f r om t he w agons Indi a l i mi t ed, w hi ch w as i ncor por at ed i n 1974.
Wagons Indi a l i mi t ed suppl i es about 90% of t he w agon r equi r ement and t he r est ar e pur chased f r om
t he open mar ket . The t ot al cost of ow ner shi p i s an i mpor t ant i ssue dur i ng t he pr ocur ement of w agons.
The pr ocur ement cost of w agon i s $40,000.
The BOXN wagons ar e mai nl y used f or car r yi n g coal and ar e f i t t ed wi t h CASNUB bogi es. CASNUB
bogi es ar e t he cr i t i cal subsyst em of t he w agon. The CASNUB bogi e consi st s of t w o cast si de f r ames and
a f l oat i ng bol st er . The bol st er i s suppor t ed on t he si de f r ames t hr ough t w o gr oups of spr i ng, w hi ch al so
i ncor por at e t he l oad pr opor t i onal f r i ct i on dampi n g. The si de f r ames of t he CASNUB bogi e ar e
connect ed by a f abr i cat ed mi l d st eel spr i ng pl ank t o mai nt ai n t he bogi e squar e. The sal i ent f eat ur es
CASNUB bogi es ar e shown i n Tabl e 1.
Tabl e 1. Sal i ent f eat ur es of CASNUB bogi e
59.1.1.1 Gauge
1676mm
Axl e l oad 20.3t . How ever al l bogi es except Casnub 22HS can be
upgr aded upt o 22.9t
Wheel Di amet er 1000 mm (New ). 956 mm ((New ) f or r et r of i t t ed
Casnub 22W
Wheel Base 2000mm
Type of Axl e Casnub 22W (M )
Bear i ng (a) Cyl i ndr i cal r ol l er bear i ng axl e box i n l i mi t ed no. on
Casnub 22W Bogi es onl y.
(b) St andar d AAR t aper ed car t r i dge bear i ng cl ass ‘ E’
sui t abl e f or 152.4x276.4mm (6” x11” ) nar r ow j aw .
289
Di st ance Bet w een
Jour nal cent r es
2260mm
Di st ance bet w een si de
bear er s
1474 mm
Type of si de bear er s
59.2 Casnub 22W
59.3 Roller Type (clearance type)
59.4 Retrofitted Casnub 22W, Casnub
22W(M), 22NLB
Const ant cont act t ype (M et al bonded r ubber pad,
housed i nsi de si de bear er housi ng).
Casnub 22HS
Spr i ng l oaded const ant cont act t ype si de bear er .
Type of pi vot Casnub 22W
IRS Type
Top Pi vot –RDSO Dr g. No. W/ BE601.
Bot t om Pi vot – RDSO Dr g. No. W/ BE602 or si mi l ar
mat i ng pr of i le i nt egr al l y cast w i t h bol st er .
59.5 Casnub 22W(M), 22NL, 22NLB,
22HS
Spher i cal t ype RDSO Dr g. No. WD85079S/ 2.
59.6 Anti rotation
features
Ant i r ot at i on l ugs have been pr ovi de bet w een bogi e
bolst er and si def r ame.
59.6.1.1 Type of brake
beam
59.7 Casnub 22W, /22NL, 22NLB and
22HS
Uni t t ype f abr icat ed br ake beam suppor t ed and
gui ded i n t he br ake beam pocket s.
Casnub22W(M)
Unit t ype cast st eel br ake beam suspended by
hanger s f r om si de f r ame br acket s.
Suspensi on det ai l s Long t r avel hel i cal spr i ngs
El ast omer i c pads On al l t ype of bogi es except Casnub
290
22Wsubsequent l y pr ovi ded i n r et r of i t ment .
The CASNUB bogi e assembl y consi st s of t he f ol l ow i ng component s:
1. Wheel set w i t h cyl indr i cal r ol l er bear i ng or w heel set w i t h car t r i dge bear ing.
2. Axl e box/ adapt er , r et ai ner bol t & si de f r ame key assembl y.
3. Si def rame w i t h f r i ct i on w ear pl at es.
4. Bolst er w i t h w ear l i ner s.
5. Spr ing pl ank, f i t bol t s & r i vet s.
6. Load bear i ng spr i ngs and snubber spr i ngs.
7. Fr ict i on shoe w edge.
8. Cent r e Pi vot ar r angement compr i si ng of cent r e pi vot , cent r e pi vot bot t om, cent r e
pi vot pi n, cent r e pi vot r et ai ner & l ocki ng ar r angement .
9. Si de bear er s.
10. Elast omer ic pads.
11. Bogi e br ake gear .
12. Br ake beam.
Rel i abi l i t y and Mai nt enance of BOXN Wagon
Indi an r ai l ways cl assi f i es t he f ai l ur es i nt o t he f ol l owi ng t hr ee cat egor i es:
1. Vi t al – causi ng l i ne f ai l ur e.
2. Essent i al – causi ng del ay t o t r af f i c.
3. Nonessent i al – causi ng no di st ur bance t o t r af f i c.
The above cl assi f i cat i on enabl es t he Indi an Rai l w ays t o f ocus on vi t al and essent i al component s
and t o st udy t hei r r el i abi l i t y and mai nt ai nabi l i t y i n ser vi ce and t ake adequat e st eps t o i mpr ove t hei r
per f or mance by modi f i cat i on or r edesi gn. The f ol l ow i ng t hr ee t ypes of mai nt enance ar e pr act i ced
f or wagons:
1. Prevent i ve mai nt enance (PM ): Pr event i ve mai nt enance i s car r i ed out af t er ever y 6000 Km f or
BOXN wagons (appr oxi mat el y 15 days).
2. Rout i ne over haul (ROH): Rout i ne over haul i s car r i ed out af t er ever y 24 mont hs. Dur i ng ROH,
t he bogi e i s di smant l ed and t he wheel s ar e dewheel ed.
3. Peri odi c Overhaul (POH): The per i odi c over haul i s car r i ed out af t er ever y 48 mont hs and
i nvol ves compl et e over haul of t he wagon. How ever , t he f i r st POH i s car r i ed out af t er 6 year s.
For t hi s r esear ch, w e l ooked at t he most cr i t i cal comp onent s (t he component s t hat cont r i but e t ow ar ds
maj or i t y of t he f ai l ur es). Tabl e 2 show s t he vi t al component s and t hei r t i met of ai l ur e di st r i but i on al ong
291
w i t h t he est i mat ed par amet er s. To mai nt ai n t he conf i dent i al i t y of t he f ai l ur e and mai nt enance dat a, w e
have used hypot het i cal dat a i n t he r est of t he paper . The obj ect i ve her e i s t o i l l ust r at e t he model s
devel oped i n t he paper .
Tabl e 2. Vi t al component s of BOXN w agons and t hei r t i met of ai l ur e di st r i but i on (λ i s t he f ai l ur e r at e, η
i s t he scal e par amet er and β i s t he shape par amet er )
S. No. Component Ti met oFai l ure
Di st ri but i on
Par amet er s
1. Wheel Wei bul l η= 52, 0000 Km, β = 4
2. Rol l er Bear i ng Wei bul l η= 250, 0000 Km, β = 3
3. Br ake Beam Wei bul l η= 160, 0000 Km, β = 4
4. Br ake Shoe Wei bul l η= 140, 0000 Km, β = 3
5. CBC Wei bul l η= 70, 0000 Km, β = 3.5
6. Panel Hat ch Wei bul l η= 38, 0000 Km, β = 4.2
7. Ai r Br ake Wei bul l η= 48, 0000 Km, β = 3.5
8. Wagon Door Exponent i al λ = 6.6 x 10
6
9. Cent r e Pi vot Wei bul l η= 55, 0000 Km, β = 3.8
Al l cr i t i cal component s except wagon door f ol l ow Wei bul l di st r i but i on. The t i met of ai l ur e di st r i but i on
of t he w agon door i s exponent i al , si nce most of t he w agon door f ai l ur es ar e caused due t o mi shandl i ng.
The t i met of ai l ur e of t he w agon i t sel f f ol l ow s exponent i al di st r i but i on w i t h mean t i me bet w een f ai l ur es
of 16000 Km.
Cal cul at i on of t he oper at i onal avai l abi li t y of Wagon
Al l t he l i f e uni t s ar e measur ed i n t er ms of Ki l omet er and t hus t he PM , ROH, POH ar e conver t ed i n
t er ms of Km. The pr event i ve mai nt enance i nt er val i s appr oxi mat el y 15 days, t hat i s af t er ever y 6000 Km,
and dur i ng PM , t he w agon i s out of ser vi ce f or 2 days (t hat i s 800 Km). Whenever , t he w agon r equi r es
cor r ect i ve mai nt enance, i t i s l i kel y t o be out of ser vi ce f or 4 days (t hat i s 1600 Km). The usage of w agon
f or ever y mont h i s 12,000 Km. Usi ng t hese dat a, t he mean t i me bet w een mai nt enance i n one year ,
(144,000 Km) i s gi ven by:
4363
6000
144000
) 144000 (
144000
·
+
·
M
MTBM
wagon
Km (12)
The Down t i me i s gi ven by:
292
1018
24 9
800 24 1600 9
) (
) ( ) (
·
+
× + ×
·
+
× + ×
·
sm
sm
wagon
T
T
T M
MPMT
T
T
MCMT T M
DT Km (13)
Usi ng, (12) and (13), we get t he oper at i onal avai l abi l i t y of t he wagon as:
8108 . 0
1018 4363
4363
·
+
·
+
·
wagon wagon
wagon
wagon
DT MTBM
MTBM
A (14)
Thus, t he oper at i onal avai l abi l i t y of t he w agon i s 81.08%.
4.3 Oper at i ng cost f or Wagon
The oper at i onal avai l abi l i t y val ue can be now used t o cal cul at e t he oper at i n g cost of t he w agon.
For t he sake of mat hemat i cal si mpl i ci t y, w e cal cul at e t he cost of ow n er shi p f or 6 year s f r om
commi ssi oni ng of t he wagon. Assume:
C
ou
= $1 per Km
Then, t he oper at i ng cost f or f i r st si x year s, at an i nt er est r at e of 6% i s gi ven i n t he f ol l ow i ng t abl e (t abl e
3):
Tabl e 3: Pr esent val ue of t he oper at i ng cost
Year PV of t he oper at i ng cost (i n $)
1
9178.868
2
8659.309
3
8169.16
4
7706.755
5
7270.523
6
6858.984
Tot al 54314.34
293
4.4 Mai nt enance cost f or Wagon
The cost of mai nt enance f or si x year s can be cal cul at ed usi ng t he equat i on (8). We make t he
f ol l ow i ng assumpt i ons:
C
cm
= Cost of cor r ect i ve mai nt enance = $800
C
pm
= Cost of pr event i ve mai nt enance = $1500
C
OH,1
= Cost of r egul ar over haul (ROH) = $8000
C
OH,2
= Cost of per i odi c over haul (POH) = $15000
Tabl e 4, show s t he pr esent val ue of t he mai nt enance cost f or t he f i r st si x year s.
Tabl e 4: Pr esent val ue of t he mai nt enance cost
Year PV of t he l ogi st i cs cost (i n $)
1
30849.06
2
36222.86
3
27455.55
4
37782.87
5
24435.34
6
28691.89
Tot al
30849.06
294
In t abl e 4, one can not i ce, cost f l uct uat i on dur i ng year 2, 4 and 6. This i s due t o ROH and POH
car r i ed out dur i ng t hat per i od.
4.5 Logi st i cs Support Cost
The l ogi st i cs suppor t cost can be est i mat ed usi ng equat i on (10). Thi s i nvol ves t he use of r enew al
pr ocess t o est i mat e t he spar es r equi r ement f or each of t he component s show n i n t abl e 2. Assumi ng, N
s
x C
s
= $ 3000, t he pr esent val ue of t he l ogi st i cs cost f or si x year s i s show n i n Tabl e 5.
Tabl e 5: Pr esent val ue of t he l ogi st i cs cost
Year PV of t he l ogi st i cs cost (i n $)
1
2830.189
2
2669.989
3
2518.858
4
2376.281
5
2241.775
6
2114.882
Tot al
14751.97
4.6 Tot al cost of ownershi p of wagon
The t ot al cost of ow ner shi p of w agon f or t he f i r st 6 year s i s obt ai ned by addi ng t h e component s
gi ven by equat i ons (6), (8) and (10). Assume t hat C
M F
= $20000. The t ot al cost of ow n er shi p, TCO
D
, f or
si x year s, i s gi ven by:
6 . 314502 $
20000 97 . 14751 6 . 185437 34 . 54314 40000
] [
,
1
, ,
·
+ + + + ·
+ + ∑ + + ·
·
MF n L
D
n
n M n O P D
C C C C C TCO
The cost of ow ner shi p i s cal cul at ed f or si x year s, w hi ch i s t he f i r st maj or over haul per i od. The above
cost can b e di vi ded by t he dur at i on, t o cal cul at e TCO per year , w hi ch t hen can be used f or compar i n g
di f f er ent conf i gur at i ons.
295
5. Deci si on Maki ng on t he Basi s of Cost of Ownershi p
In t hi s sect i on, w e di scuss how t he cost of ow ner shi p der i ved i n t he pr evi ous sect i on can be used f or
pur chasi ng deci si ons. The f oll ow i ng t w o appr oaches can be used f or deci si on maki ng.
5.1 Deci si on Maki ng Based on TCO as onl y Cri t er i a
If t he cost of ow ner shi p i s t he onl y cr i t er i a on w hi ch t he pur chasi ng deci si on i s based on t hen,
t he al t er nat i ve w i t h mi ni mum t ot al cost of ow ner shi p per per i od (per annum). That i s, i f t her e ar e n
al t er nat i ves such t hat :
TCO
a,i
= Tot al cost of ow ner shi p per annum f or t he i
t h
al t er nat i ve.
Then, t he al t er nat i ve, m, such t hat :
} TCO ,..., TCO , Min{TCO TCO
n a, a,2 a,1 m a,
· (15)
5.2 Deci si on Maki ng Based on TCO as one of t he cri t er i a
In many si t uat i ons, t he pur chasi ng deci si on i s made usi ng mul t i pl e cr i t er i a, i ncl udi ng TCO as o ne
of t he cr i t er i a. In such cases, one can use mul t i cr i t er i a deci si on maki ng t echni qu es t o choose t he best
al t er nat i ve. In t hi s paper , we suggest anal yt i c hi er ar chy pr ocess (AHP) f or choosi ng t he best al t er nat i ve
(Saat y, 1980). AHP i s a mul t i cr i t er i a d eci si on maki ng t echni que w hi ch can be used t o choose best
al t er nat i ve among number of al t er nat i ves. Let us assume t hat t her e ar e M al t er nat i ves and N deci si on
cr i t er i a. Let a
i j
denot e t he wei ght of t he i
t h
al t er nat i ve on j
t h
cr i t er i a. Let W
j
be t he w ei ght f or cr i t er i a j .
Then, t he deci si on pr obl em can be def i ned usi ng t he f ol l ow i ng mat r i x (Tr i ant aphyl l ou et al 1995)
Cr i t er i a
Al t er nat i ve
1 2 3 N
W
1
W
2
W
3
… W
N
1 a
11
a
12
a
13
… a
1N
2 a
21
a
22
a
23
… a
2N
.
.
…
…
296
. …
M a
M 1
a
M 2
a
M 3
a
M N
Usi ng t he above dat a AHP f i nds t he over al l i mpor t ance of t he al t er nat i ve and chooses t he one w i t h
maxi mum wei ght (i nt er est ed r eader may r ef er t o Saat y, 1980).
6. Concl usi ons
The mai n obj ect i ves of t hi s pap er ar e t o devel op f r amew or k f or est i mat i on of cost of ow ner shi p of
capi t al asset s and t o devel op mat hemat i cal model s f or est i mat i on of var i ous cost el ement s w i t hi n t he
cost of ow ner shi p. The i nno vat i ve appr oach used i n t hi s paper i s t he use of oper at i onal avai l abi l i t y t o
est i mat e t he oper at i ng cost . M ost of t he model use cal endar t i me t o est i mat e t he oper at i n g t i me. The
use of cal endar t i me may be appr opr i at e f or cer t ai n el ement s of oper at i ng cost such as l abor cost s, t he
var i abl e cost s su ch as ener gy consumed w oul d depend on t he oper at i onal avai l abi l i t y of t he syst em. The
model s devel oped i n t he paper ar e used i l l ust r at ed usi ng t he w agons used i n t he Indi an r ai l w ays.
Al t hough, t he dat a used i n t he paper ar e modi f i ed, i t capt ur es t he i mpact of cost of ow ner shi p. In t he
exampl e, w e have show n t hat t he cost of ow ner shi p of w agon f or si x year s i s al most 8 t i mes i t s i ni t i al
pr ocur ement pr i ce. Th e mai n ai m of t hi s paper i s t o show t he si gni f i cance and t ot al cost of ow ner shi p
compar ed t o t he pr ocur ement pr i ce and t hus t o pr ove t hat al l pr ocur ement deci si ons must be based on
t ot al cost of ow ner shi p and not on t he basi s of pr ocur ement cost .
Ref er ences
Anon. (2002), Pr ocur ement of wagons by Indi an Rai l ways, Repor t Number 9A, Indi an Rai l ways.
Bl anchar d, B. S. (1986), Logi st i cs Engi neer i ng and M anagement , 3
r d
Edi t i on, Pr ent i ce Hal l , New Jer sey.
Bl anchar d, B. S. (1988), Syst ems Engi neer i ng and M anagement , 2
nd
Edi t i on, John Wi l ey, New Yor k.
Bl anchar d, B. S. and Fabr ycky, W. J. (1990), Syst ems Engi neer i ng and Anal ysi s, 2
nd
Edi t i on, Pr ent i ce Hal l ,
New Jer sey.
Di nesh Kumar , U., Cr ocker , J, Knezevi c, J and El Har am. (2000), Rel i abi l i t y, M ai nt enance and Logi st i cs
Suppor t – A Li f e Cycl e Appr oach, Kl uw er Academi c Publ i sher s, Bost on.
Hampt on, T. (2004), The Cost of Ow ner shi p, ENR: Engi neer i ng News Recor d, Vol . 253, 7, 2429.
Har vey, G. (1976), Li f e Cycl e Cost i ng: A Revi ew of t he Techni que, M anagement Account i ng, 343347.
Humphr i es, J. (2004), Opt i mi zi ng Tot al Cost of Owner shi p, Pl ant Engi neer s, Vol. 58, 7, 2327.
Ross, S. (2000), Int r oduct i on t o Pr obabi l i t y M odel s, Academi c Pr ess, New Yor k.
297
Saat y, T. L. (1980), The Anal yt i c Hi er ar chy Pr ocess: Pl anni ng, Pr i or i t y Set t i ng and Resour ce Al l ocat i on,
M cGr aw Hi l l .
Sher i f , Y. S., and Kol ar i k, W. J. (1981), Li f e Cycl e Cost i ng: Concept and Pr act i ce, OM EGA, 9,.3, 287296.
Tayl or , W. B. (1981), The Use of Li f e Cycl e Cost i ng i n Acqui r i ng Physi cal Asset , Long Range Pl anni ng, 14,
3243.
Tr i ant aphyl l ou, E., and M ann, S. (1995), Usi ng Anal yt i cal Hi er ar chy Pr ocess i n Deci si on M aki ng i n
Engi neer i ng Appl i cat i ons: Some Chal l enges, Int er nat i onal Jour nal of Indust r i al Engi neer i ng:
Appl i cat i ons and Pr act i ce, Vol . 2, No. 1, 3544
1. Reliability Maintenance and Logistic Support  Introduction
2
Chapter 1 Reliability Maintenance and Logistic Support  Introduction
All the business of war, and indeed all the business of life, is to endeavour to find out what you don t know from what you do. Duke of Wellington
1.1.
INTRODUCTION
Ever since the Industrial Revolution began some 2½ centuries ago, customers have demanded better, cheaper, faster, more for less, through greater reliability, maintainability and supportability (RMS). As soon as people set themselves up in business to provide products for others and not just for themselves, their customers have always wanted to make sure they were not being exploited and that they were getting value for money and products that would be fit for purpose. Today’s customers are no different. All that has changed is that the companies have grown bigger, the products have become more sophisticated, complex and expensive and, the customers have become more demanding and even less trusting. As in all forms of evolution, the Red Queen Syndrome (Lewis, C. 1971, Matt, R., 1993) is forever present – in business, as in all things, you simply have to keep running faster to stand still. No matter how good you make something, it will never remain good enough for long. Operators want infinite performance, at zero lifecycle cost, with 100% availability from the day they take to delivery to the day they dispose of it. It is the task of the designer/manufacturer/supplier/producer to get as near as possible to these extremes, or, at the very least, nearer than
1. Reliability Maintenance and Logistic Support  Introduction
3
their competitors. In many cases, however, it is not simply sufficient to tell the (potential) customer how well they have met these requirements, rather, they will be required to produce demonstrable evidence to substantiate these claims. In the following pages, we hope to provide you with the techniques and methodologies that will enable you to do this and, through practical examples, explain how they can be used. The success of any business depends on the effectiveness of the process and the product that business produces. Every product in this world is made to perform a function and every customer/user would like her product to maintain its functionality until has fulfilled its purpose or, failing that, for as long as possible. If this can be done with the minimum of maintenance but, when there is a need for maintenance, that this can be done in the minimum time, with the minimum of disruption to the operation requiring the minimum of support and expenditure then so much the better. As the consumer’s awareness of, and demand for, quality, reliability and, availability increases, so too does the pressure on industry to produce products, which meet these demands. Industries, over the years, have placed great importance on engineering excellence, although some might prefer to use the word “hubris”. Many of those which have survived, however, have done so by manufacturing highly reliable products, driven by the market and the expectations of their customers. The operational phase of complex equipment like aircraft, rockets, nuclear submarines, trains, buses, cars and computers is like an orchestra, many individuals, in many departments doing a set of interconnected activities to achieve maximum effectiveness. Behind all of these operations are certain inherent characteristics (design parameters) of the product that plays a crucial role in the overall success of the product. Three such characteristics are reliability, maintainability and supportability, together we call them RMS. All these three characteristics are crucial for any operation. Billions of dollars are spent by commercial and military operators every year as a direct consequence of the unreliability, lack of maintainability and poor supportability of the systems they are expected to operate. Modern industrial systems consist of complex and highly sophisticated elements, but at the same time, users’ expectations regarding trouble free operation is ever present and even increasing. A Boeing 777 has over 300,000 unique parts within a total of around 6 million parts (half of them are nuts, bolts and rivets). Successfully operating, maintaining and supporting such a complex system demands integrated tools, procedures and techniques. Failure to meet high reliability, maintainability and supportability can have costly and farreaching effects. Losing the services
1. Reliability Maintenance and Logistic Support  Introduction
4
of airliners, such as the Boeing 747, can cost as high as $ 300,000 per day in forfeited revenue alone. Failure to dispatch a commercial flight on time or its cancellation is not only connected to the cost of correcting the failure, but also to the extra crew costs, additional passenger handling and loss of passenger revenue. Consequently, this will have an impact on the competitiveness, profitability and market share of the airline concerned. 'Aircraft on Ground' is probably the most dreaded phrase in the commercial airlines’ vocabulary. And, although the costs and implications may be different, it is no more popular with military operators. Costs per minute delay for different aircraft type are shown in Figure 1.1. Here the delay costs are attributable to labour charges, airport fees, air traffic control costs, rescheduling costs, passenger costs (food, accommodation, transport and payoffs).
Figure 1.1 Aircraft delay cost per minute Industries have learned from past experience and through cutting edge research how to make their products safe and reliable. NASA, Boeing, Airbus, Lockheed Martin, RollsRoyce, General Electric, Pratt and Whitney, and many, many more, are producing extremely reliable products. For example, over 25% of the jetliners in US have been in service for over 20 years and more than 500 over 25 years, nearing or exceeding their original design life (Lam, M., 1995). The important message is that these aircraft are still capable of maintaining their airworthiness; they are still safe and reliable. But, we cannot be complacent, even the best of organisations can have their bad days. The losses of the Challenger Space Shuttle in 1986, and Apollo 13 are still very fresh in many of our memories. Customers’ requirements generally exceed the capabilities of the producers. Occasionally, these go beyond what is practically, and sometimes even theoretically, possible. An example of this could be the
1. Reliability Maintenance and Logistic Support  Introduction
5
new reliability requirement, maintenance and failure free operating period, (Hockley et al 1996, Dinesh Kumar et al, 1999, 2000). High reliability is certainly a desirable function, but so to is maintainability and excellent logistic support. It is only through all three that the lifecycle cost can be driven down whilst the level of availability is driven up. Combat aircraft are expensive and so are their crews, so no operator wants to lose either. At the same time, deploying large ground forces to maintain and support them is also expensive and, potentially hazardous. It is therefore not surprising that the operators are looking to the manufacturers to produce aircraft so reliable that they can go for weeks without any maintenance. The question is, however, can we achieve the necessary level of reliability, with sufficient confidence, at an affordable price, to meet this requirement?. Recent projects such as the Ultra Reliable Aircraft (URA) and Future Offensive Air Systems (FOAS) place a new dimension to the reliability requirement. The operators/users would like to have Maintenance Free Operating Periods (MFOP), during which the probability that the system will need restorative maintenance is very low. Between each of these periods, sufficient maintenance will done to ensure the system will survive the next MFOP with the same probability. Only time will tell whether this policy becomes adopted but, there is no doubt that the days of the MTBF (mean time between failures) and its inverse, the [constant] failure rate are surely numbered. Science, mathematics and probability theory are slowly finding their way into the aftermarket business and with them will come the need for better educated people who understand these new concepts, techniques and methodologies. And, it will not just affect military aircraft, buyers of all manufactured products will demand greater value for money, at the time of purchase, of course, but more than that they will expect it throughout its life. Manufacturers who have relied on unreliability will need to rethink their policies, processes and finances.
1.2.
THE LIFE CYCLE OF A SYSTEM
Fundamental to any engineering design practice is an understanding of the cycle, which the product goes through during its life. The life cycle begins at the moment when an idea of a new system is born and finishes when the system is safely disposed. In other words, the life cycle begins with the initial identification of the needs and requirements and extends through planning, research, design, production, evaluation, operation, maintenance, support and its ultimate phase out (Figure 1.2).
The main reason for the need of a new system could be a new function to be performed (that is there is a new market demand for a product with the specified function) or a deficiency of the present system. 2. on the other hand. Manufacturers who specialise in military hardware will often be approached. 5. Poor reliability. Airbus Industries. that there was a sufficient market need for a very large aircraft that could carry well in excess of 500 passengers.1. The output from this stage is fed to the preliminary design stage. either directly or through an advertised “invitation to tender” to discuss the latest defence requirement. This analysis would then lead to conceptual system design alternatives. 1995). decided. based on their extensive market research. 4. both friend and foe. Inadequate performance. High maintenance and support costs. Reliability Maintenance and Logistic Support . It will be many years before we will know whether either of these aircraft will get off the ground and very much longer to see if they prove a business success for their manufacturers. at least across the Pacific from Tokyo to Los Angeles and possibly even nonstop between London and Sydney. For most other manufacturers. The flow of the conceptual system design process is illustrated in Figure 1. Low sales figures and hence low profits. 5. Functional deficiencies. The deficiencies could be in the form of: 1. 3. The first step in the conceptual design phase is to analyse the functional need or deficiency and translate it into a more specific set of qualitative and quantitative requirements. The first process then is a set of tasks performed to identify the needs and requirements for a new system and transform them into its technically meaningful definition. The .Introduction 6 Needs and Requirements Design Production or Construction Use Retirement Conceptual design Preliminary design Detailed design Manufacture Assembly Operation Maintenance Support Figure 1. and that would include all of the concepts identified as practical in the URA research project. The UK MoD approached BAE Systems to bring together a consortium (including representatives of the MoD and RAF) for an air system that would outperform all existing offensive systems. it is generally up to them to identify a (potential) market need and decide whether they can meet that need in a profitable way. Inadequate attributes.3 (D Verma and J Knezevic.2 Life cycle of the system.
however. for example. Reliability. It is during this stage that safety. 3. maintenance functions. It is at this time that the concepts are brought down to earth out of the “blue sky”. It was largely a result of this activity that the concepts of the MFOP and the uninhabited combat air vehicle (UCAV) were born. Reliability Maintenance and Logistic Support . During the production/construction process the system is physically created in accordance with the design definition. The main tasks performed during this process are 1. equipment. Design is the most important and crucial stage in the product life cycle. 2. and 4. The input characteristics of the production process are the raw material. Needs and Requirements Needs Analysis & Requirements Definition Synthesis of conceptual system design alternatives Analysis of Conceptual System Design Alternatives Evaluation of Conceptual System Design Alternatives Figure 1. energy.1. maintainability and supportability depend on the design and are the main drivers of the operational availability and costs. facilities and other ingredients needed for the production/construction of the new . maintenance and support plans can be decided. System prototype test and evaluation. impractical. Manufacture/Production/Test of prime system elements. from these.Introduction 7 conceptual design stage is the best time for incorporating reliability. Quality Assurance. System Modification. System assessment. Groups will be required to put these ideals into reality possibly via technical development programs or abandon them until the next time. Development of system prototype. 1991). reliability and maintainability demonstrations can be performed and. 2. maintainability and supportability considerations. various integrated project teams with representatives of the users. The production/construction process is a set of tasks performed in order to transform the full technical definition of the new system into its physical existence. In the case of FOAS.3 Conceptual system design process The main tasks during the preliminary design stage are system functional analysis such as operational functions. suppliers and even academia will drawn together to come up with new ideas and set targets. The main tasks performed during the detailed design stage 1. Development of system/product design. and 3. allocations of performance and effectiveness factors and the allocation of system support requirement (Blanchard.
a tyre or bearing wears. a tyre punctures. Similarly. Inspecting the amount of tread on the tyres at regular intervals. state of failure. a failure of a system is any event or collection of events that causes the system to lose its functionability where functionability is the inherent characteristic of a product related to its ability to perform a specified function according to the specified requirements under the specified operating conditions. Reliability Maintenance and Logistic Support . wherever possible. (Knezevic 1993) Thus a system. may be deferred to a time which is more convenient to the operator. to mitigate against such events. the transition between these states is effectively instantaneous. 1. Kalman Filtering) on the specific fuel consumption can alert the user to imminent onset of failure. In these cases. the transition is gradual.g. whose failure may have serious or catastrophic consequences. safe in the knowledge that there is an acceptably high probability that the system will continue operating safely for a certain length of time. in many other cases. The output characteristics are the full physical existence of the functional system. The recovery of the failed item.1. Cars are fitted with dual braking systems. With many highly complex systems. Essentially. a crack propagates across a disc. it is possible to have a failure of a component without a failure of the system. If one of the flight control computers on an aircraft fails. some form of health monitoring may allow the user to take preventative measures. In these circumstances. In many cases. There is insufficient time to detect the onset or prevent the consequences. erosion.Introduction 8 system. However. scanning the lubricating oil for excessive debris. corrosion or any of the other visible or physically detectable signs that might cause the component to become nonfunctionable. a blade “creeps” or the performance starts to drop off. a windscreen shatters. or indeed. measures are taken. deterioration. via a maintenance action. its functions will instantly and automatically be taken over by . a blade breaks. can only be in one of two states: state of functioning or. a transistor blows. any one of the many forms of nondestructive testing may be used (as appropriate) on components that have been exposed during the recovery of their parent component to check for damage. aircraft with (at least) triple hydraulic systems and numerous other instances of redundancy.3. failure has come to mean many things to many people. CONCEPT OF FAILURE As with so many words in the English language. any component within it. boroscope inspection to look for cracks or using some form trending (e.
that aircraft would not be permitted to take off until it was fitted with two functionable engines. the computer is not only able to detect these sectors but it will mark them as unusable and avoid writing any data to them. uninterrupted to its next scheduled destination. Again. Airbus 330 and Boeing 777). it is not uncommon for small sectors of these discs to become unusable. engines turn or passengers swim). So there is both fault tolerance and redundancy although the latter is usually at the discretion of the user. During the life of the PC. the builtin test software of the computer is able to provide a level of fault tolerance which is often totally invisible to the user. if that program or data has been backed up to another medium. Unfortunately. Provided the sector did not hold the file access table (FAT) or key system’s files. although with special software. “rotation” or “weightoffwheels”. the aircraft is put in for scheduled maintenance. it should be possible to restore the full capacity of the system usually with a level of manual intervention. this will no longer be accessible. it would not then be permitted to takeoff again until that engine has been returned to a state of functioning (except under very exceptional circumstances). . having landed. Even under these circumstances.g. Reliability Maintenance and Logistic Support . the system (aircraft) has a limited level of fault/failure tolerance. once it has landed. neither engine is truly redundant but. Part of the certification process requires a practical demonstration that a fully loaded aircraft can takeoff safely even if one of those engines fails at the most critical time.Introduction 9 one of the other computers. either until another computer fails or.1. It also requires an aircraft that has “lost” an engine to fly to immediately divert to a landing site that is within this flying time. engines. a change in the airworthiness rules has allowed them to fly for extended periods following the inflight shutdown of one of the engines. Most commercial airliners are fitted with two. further flights may be permitted. generally referred to ETOPS (which officially stands for extended twin operations over sea or. This defines the maximum distance (usually expressed in minutes of flying time) the aircraft can be from a suitable landing site at any time during the flight. The flight will generally be allowed to continue. even though the aircraft can fly with one engine out of service. In this case. it may be possible to recover some of it. However. if there was already data on these sectors before they become unusable. or more. With the latest large twins (e. Depending on the level of redundancy and regulations/certification. Thus. Most personal computers (PC) come complete with a “hard disc”. at least until the whole disc crashes or the fault affects a critical part of a program or data. unofficially.
The intention is to provide an understanding of the main concepts in probability theory that can be applied to problems in reliability. PROBABILITY TERMS AND DEFINITIONS In this section those elements essential for understanding the rudiments of elementary probability theory will be discussed and defined in a general . 10 Chapter 2 Probability Theory We do not know how to predict what would happen in any given circumstances. which are discussed in the following chapters. This chapter is not intended to a rigorous treatment of allrelevant theorems and proofs. or trial is uncertain and a prediction has to be made.2. Probability theory is applicable to everyday life situations where the outcome of a repeated process. that the only thing that can be predicted is the probability of different events Richard Feynman Probability theory plays a leading role in modern science in spite of the fact that it was initially developed as a tool that could be used for guessing the outcome of some games of chance. experiment. maintenance and logistic support.4. and we believe now that it is possible. In order to apply probability to everyday engineering practice it is necessary to learn the terminology. 2. test. definitions and rules of probability theory.
Experiment An experiment is a welldefined act or process that leads to a single welldefined outcome.c. Figure 2. for elementary events that are possible outcomes of the experiment under consideration. Be capable of being described.b. together with illustrative examples related to engineering practice. Elementary event An elementary event is every separate outcome of an experiment. and small letters. since every experiment must have only one outcome..1 Graphical Representation of an Experiment and its outcomes. so that the observer knows when it occurs. Every experiment must: 1. Experiment Figure 2. From the definition of an experiment. a.. 2. Sample space The set of all possible distinct outcomes for an experiment is called the sample space for that experiment.. Probability Theory 11 manner.1 illustrates the concept of random experiments. Most frequently in the literature the symbol S is used to represent the sample space. To facilitate the discussion some relevant terms and their definitions are introduced. so that the set of all possible outcomes can be specified. Have one and only one outcome. it is possible to conclude that the total number of elementary events is equal to the total number of possible outcomes.2. The set S may .
Figure 2. For example. then the elementary event is the speed measured. whereas the sample space consists of all the different speeds one might possibly record. 2.5.2 is a graphical presentation of the sample space.2 Graphical Presentation of the Sample Space Event Event is a subset of the sample space. C (between 50 and 70 km/h) and D (above 70 km/h). that is. All speed events could be classified in. four different speed groups: A (less than 30 km/h). All other rules and relations are derived from them. then the event B is said to have occurred. say. a collection of elementary events. it is necessary to assign some measure that will indicate the chances of . ELEMENTARY THEORY OF PROBABILITY The theory of probability is developed from axioms proposed by the Russian mathematician Kolmogrov.2. Figure 2. are usually used for denoting events. B (between 30 and 50 km/h). Capital letters A.5. 2. If the measured speed of the passing car is. In practice this means that its elements have been defined together with several axioms which govern their relations. Probability Theory 12 contain either a finite or an infinite number of elementary events. B. if the experiment performed is measuring the speed of passing cars at a specific road junction. C.1 Axioms of Probability In cases where the outcome of an experiment is uncertain. …. say 35 km/h.
2. and that the probability of the sure event S. Formally.. The probability of the union of mutually exclusive events is the sum of their probabilities.. that is P( A1 ∪ A2 . is always 1. the probability function is defined as: A function which associates with each event A. (probability of the sample space) 3. Probability Theory 13 occurrence of a particular event. is called the probability function ..3.). Figure 2. ( P(A) denotes the probability of event A).. such that the following axioms are true: 1. The function which associates each event A in the sample space S.3 Graphical representation of probability function.2. Such a measure of events is called the probability of the event and symbolised by P(. if A1 and A2 are any two mutually exclusive events (that is. the occurrence of one event implies the nonoccurrence of the other) in the .the probability of that event. P(S) = 1. Furthermore. P(A). P(A) > 0 for every event A. this definition states that each event A is paired with a nonnegative number. A graphical representation of the probability function is given in Figure 2. probability P(A). ∪ An ) = P( A1 ) + P( A2 ) + . + P( An ) In essence. with the probability measure P(A). the probability of event A. or P(S). a real number.
φ. then the probability of A is less than or equal to the probability of B: P( A) ≤ P( B) (2. is simply the sum of their two probabilities. the probability of A.2) P(φ ) = 0 (2. Probability Theory 14 sample space.1) b) The probability of any event must lie between zero and one inclusive: 0 ≤ P( A) ≤ 1 c) The probability of an empty or impossible event.3) d) If occurrence of an event A implies that an event B occurs. (2.5. then P( A ∪ B) = P( A) + P( B) one: (2. the probability of B. then their probabilities must add up to .2 Rules of Probability The following elementary rules of probability are directly deduced from the original three axioms. P( A1 ) + P( A2 ) . is zero. the probability of the complementary event. and also the probability that both occur must be known.5) f) If A and B are mutually exclusive events. written A' .4) e) In order to find the probability that A or B or both occur. 2. so that the event class A is a subset of event class B. so that P( A ∩ B) = 0 . using the set theory: a) For any event A. the probability of their union P( A1 ∪ A2 ) .2.6) g) If n events form a partition of S. is given by P( A' ) = 1 − P( A) (2. thus: P ( A ∪ B ) = P ( A) + P ( B ) − P ( A ∩ B ) (2.
7) 2. There is nothing to restrict any given elementary event from the sample space from qualifying for two or more events. the joint event is A ∩ B . denoted by P( B A) . is defined as: P( B A) = P(A ∩ B) . both A and B events occur when A ∩ B occurs. + P( A n ) = ∑ P( Ai ) = 1 i =1 (2.5.3 Joint Events Any event that is an intersection of two or more events is a joint event..4 Conditional Probability If A and B are events in a sample space which consists of a finite number of elementary events. Since a member of A ∩ B must be a member of set A. and also of set B. Probability Theory 15 n P( A1 ) + P( A2 ) + .2. provided that those events are not mutually exclusive.5. Provided that the elements of set S are all equally likely to occur. the probability of the joint event could be found in the following way: P( A ∩ B ) = number of elementary events in A ∩ B total number of elementary events 2. Thus. given the event A and the event B..8) . the conditional probability of the event B given that the event A has already occurred. P(A) P ( A) > 0 (2.
and if B is subset of ( A1 ∪ A2 ∪K∪ A N ) . the relative frequency of occurrence of an event. A N ) represents the partition of the sample space (N mutually exclusive events). then the probability that the relative frequency of occurrence of A. A statement of probability tells us what to expect about the relative frequency of occurrence. as illustrated in Figure 2. given A) (The probability of A. which can be stated as follows: If ( A1 . should approach the probability of this event. and is now wellknown as Bernoulli's theorem: If the probability of occurrence of an event A is p. given that enough observations are made. then P ( Ai  B ) = P(BA i )P(A i ) P(BA 1 )P(A 1 ) +K+ P(BA i )P(A i ) +K+ P(BA N )P(A N ) (2. if independent trials are made at random over an indefinitely long sequence.K . For any two events A and B. In the long run.6. Probability Theory Figure 2.9) 2. because it does not make sense to consider the probability of B given A if event A is impossible.4. (defined as f ( A) = N ( A) n ) differs . there are two conditional probabilities that may be calculated: P(A ∩ B) P(A ∩ B) P( B A) = and P ( A B) = P(A) P(B) (The probability of B. and if n trials are made independently and under the same conditions.2.4 Graphical Presentation of the Bayes Theorem 16 The conditional probability symbol. This principle was first formulated and proved by James Bernoulli in the early eighteenth century. A2 . It is necessary to satisfy the condition that P(A)>0. is read as the probability of B given A. PROBABILITY AND EXPERIMENTAL DATA The classical approach to probability estimation is based on the relative frequency of the occurrence of that event. P ( B A) . given B) One of the important application of conditional probability is due to Bayes theorem. say A.
2.7.60 Coupe 0. the n proportion actually observed might be any number between 0 and 1. This does not mean that the proportion of N ( A) occurrences among any n trial must be p. given more and more trials. imagine a sample space of all Ford cars produced. That is. approaches zero as the number of trials grows indefinitely large. and suppose that they form a partition of the sample space S. The corresponding set of probabilities. is a probability distribution. 2. Although it is true that the relative frequency of occurrence of any event is exactly equal to the probability of occurrence of any event only for an infinite number of independent trials.00 All events other than those listed have probabilities of zero . The probability distribution might be: Event P Saloon 0.K . there is very good reason to expect the observed relative frequency to be quite close to the probability because the rate of convergence of the two is very rapid.09 Total 1. However.31 Estate 0. As a simple example of a probability distribution. P ( A2 ). Nevertheless.K . P( N ( A) n) − p > s) → 0. they are mutually exclusive and exhaustive. the main drawback of the relative frequency approach is that it assumes that all events are equally likely (equally probable). P( A1 ).10) where s is some arbitrarily small positive number. An . the relative frequency of f ( A) occurrences may be expected to become closer and closer to p. A2 . as n → ∞ (2. That is. however small. PROBABILITY DISTRIBUTION Consider the set of events A1 . this point must not be over stressed. An illustrative presentation of the concept of probability distribution is shown in Figure 2.5. A car selected at random is classified as a saloon or coupe or estate. Probability Theory 17 from p by any amount. P( An ) . Even with relatively small number of trials.
an pn Probability Distribution 0 p1 p2 . b. it is possible to define the event A to be the subset of S consisting of all sample points 'a' to which the random variable X assigns the number r.. such as X. So. the event A has a probability p = P( A) . A = (a : X (a ) = r ) . for practical applications. a random variable is a realvalued function defined in a sample space. Probability Theory 18 Sample Space (S) a1 a2 an } S P a1 p1 a2 p2 . Y and Z. Usually it is denoted with capital letters.6 If X is a random variable and r is a fixed real number. such as x. the symbol P( X = r ) represents the probability function of a random variable.pn 1 Figure 2. Outcomes of experiments may be expressed in numerical and nonnumerical terms. In other words. The symbol p can be interpreted. z. p = P( X = r ) .2. and so on. . y. On the other hand. it is necessary to assign a numerical value to each possible elementary event in a sample space S. whereas small letters.5 Graphical representation of Probability Distribution 2. c. as the probability that the random variable X takes on the value r. . pi . Thus. see Figure 2. In order to compare and analyse them it is much more convenient to deal with numerical terms. The function that achieves this is known as the random variable. Even if the elementary events themselves are already expressed in terms of numbers.8. a. generally. RANDOM VARIABLE A function that assigns a number (usually a real number) to each sample point in the sample space S is a random variable. are used to denote particular values of random variables. it is possible to reassign a unique real number to each elementary event.
6 Graphical Representation of Random Variable 19 Therefore. by using the random variable it is possible to assign probabilities to real numbers. There are very many situations where the random variable X can assume only a particular finite or countably infinite set of values. which the random variables can assume. can be classified as discrete or continuous.2. F(t). That is. is called the cumulative distribution function. The main characteristics. similarities and differences for both types will be briefly described below. takes value less than or equal to certain value 'x'.8.1 Types of random variables Depending on the values. Probability Theory Figure 2. . it is said to be a discrete random variable. random variables.7. P[X ≤ x] = F(x) Figure 2. as shown in Figure 2.7 Relationship between probability function and a random variable 2. Discrete random variables If the random variable X can assume only a particular finite or countably infinite set of values. that is. although the original probabilities were only defined for events of the set S. the possible values of X are finite in number or they are infinite in number but can be put in a onetoone correspondence with a set of real number. The probability that the random variable X.
which consists of recording the temperature of a cooling liquid of an engine in the area of the thermostat at a given time. we can determine values 75. Probability Theory Continuous random variables 20 If the random variable X can assume any value from a finite or an infinite set of values.9651. if one of the finite set of values is 75. { } . it could be said that the probability distribution of the random variable is a set of pairs. maintenance and logistic support as the random variables time to failure. i = 1. THE PROBABILITY DISTRIBUTION OF RANDOM VARIABLE Taking into account the concept of the probability distribution and the concept of the random variable.9. a set which contains an infinite (and uncountable) number of values. which means that our measuring device allows us to record the temperature to any number of decimal points. What is being demonstrated here is that the possible values of X consist of the set of real numbers.8. Let us consider an experiment. For example. P ( X = ri ). Suppose that we can measure the temperature exactly. 75. n as shown in Figure 2.965. it is said to be a continuous random variable. time to repair and the logistic delay time are continuous random variables.2. If X is the temperature reading. and so on. 2. Continuous random variables have enormous utility in reliability. ri . it is not possible for us to specify a finite or countably infinite set of values. which are also possible values of X.9652.
2. where a specific mathematical functions exist from which the probability of any value or interval of values can be calculated. If the number of possible values is small. 2. In the extreme case where we have an infinite number of possible values (for example.8 Probability Distribution of a Random Variable The easiest way to present this set is to make a list of all its members. the probability density function. it is clearly impossible to make a listing. there are other methods that could be used for specifying a probability distribution of a random variable: a) Functional method. In this particular case. it is easy to specify a probability distribution. Probability Theory 21 Figure 2. The most frequently used functions for the description of probability distribution of a random variable are the probability mass function. a function is a relation where each member of the domain is paired with one member of the range. all real numbers between zero and one). and the cumulative distribution function. Fortunately.9. the relation between numerical values which random variables can have and their probabilities will be considered. a listing may become very difficult. where the entire distribution is represented through one or more parameters known as summary measures. if there are a large number of possible values. On the other hand. Each of these will be analysed and defined in the remainder of this chapter. Probability mass function .1 Functional Method By definition. b) Parametric method.
the probability that the random variables take on an exact value. Thus. a probability distribution can be represented by the set of pairs of values (ai . consider the example of an infinite set for a specific type of car.K . Given that a discrete random variable takes on only n different values. P ( X = ai ) ≥ 0 for i = 1. a probability mass function. However. What is the probability that a car selected at random will have exactly 16 litres of fuel? This could be considered as an event that is defined by the interval of values between. say 15.9 Probability Mass Function Probability density function In the previous section.11) In practice this means that the probability of each value that X can take must be nonnegative and the sum of the probabilities must be 1. a2 .K . n ∑ P ( X = ai ) = 1 i =1 n (2.75 and 16. where i = 1.9. places a mass of probability pi at the point of xi on the Xaxis. or 15.2. discrete random variables were discussed in terms of probabilities P(X =x). where i is not exactly .5. The advantage of such a graph over a listing is the ease of comprehension and a better provision of a notion for the nature of the probability distribution. where the volume of the fuel in the fuel tank is measured with only some degree of accuracy.2. which is usually denoted as PMF(. a n . pi ) . pi = P( X = xi ) . n . say a1 . or any other interval ± 16 × 0. X. Figure 2. 2.) must satisfy the following two conditions: 1. Probability Theory 22 This function is related to a discrete random variable and it represents the probability that the discrete random variable.25. Thus. the corresponding PMF(.5 and 16.K .1i .) .2. as shown in Figure 2. will take one specific value xi .
the probability distribution of a continuous random variable can be represented by its Probability Density Function. is defined as: f ( x ) = lim P ( x ≤ X ≤ x + ∆ ( x )) ∆x ∆ ( x )→ 0 (2. which is defined in the following way: P ( a ≤ X ≤ b) = ∫ f ( x ) dx a b (2. A curve for a hypothetical distribution is shown in Figure 2. we deal with the socalled probability density of X at 'a'. the probability of exactly 16 litres is.12) As a consequence. In general.13) A fully defined probability density function must satisfy the following two requirements: f ( x) ≥ 0 +∞ for all x −∞ ∫ f ( x)dx = 1 The PDF is always represented as a smooth curve drawn above the horizontal axis. the probabilities of a continuous random variable can be discussed only for intervals of X values. the smaller the probability.10 where the two points a and b on the horizontal axis represent limits which define an interval. in effect. the occurrence of any exact value of X may be regarded as having zero probability. Thus.2. Probability Theory 23 zero. f (x) . when ∆( x ) approaches zero. zero. instead of the probability that X takes on a specific value. for continuous random variables. which represents the probability that the random variable will take values within the interval x ≤ X ≤ x + ∆( x ) . . In general. Since the smaller the interval. say 'a'. PDF. symbolised by f (a ) . The Probability Density Function. which represents the possible values of the random variable X.
The general symbol F (x ) is sometimes used to represent the function relating the various values of X to the corresponding cumulative probabilities.10 Probability Density Function for a Hypothetical Distribution The shaded portion between 'a' and 'b' represents the probability that X takes on a value between the limits 'a' and 'b'. 0 ≤ F ( x) ≤ 1 2. and it must satisfy certain mathematical properties. F ( ∞) = 1 and F ( −∞) = 0 . if a < b.14) The symbol F (a ) denotes the particular probability for the interval X ≤ a . the most important of which are: 1. F ( a ) ≤ F ( b) 3. CDF.2. Cumulative distribution function The probability that a random variable X takes on a value at or below a given number 'a' is often written as: F (a) = P( X ≤ a) (2. Probability Theory 24 Figure 2. This function is called the Cumulative Distribution Function.
11 Cumulative Distribution Function for Discrete Variable 1 25 F(x) F(a) 0 a x Figure 2. Two general groups of such characteristics applicable to any type of distribution are: a) Measures of central tendency (or location) which indicate the typical or the average value of the random variable.16) Hypothetical cumulative distribution functions for both types of random variable are given in Figures 2.12 Cumulative Distribution Function for Continuous Variable The symbol F ( x ) can be used to represent the cumulative probability that X is less than or equal to x. 2. . It is defined as: F (a ) = ∑ P( X = xi ) i =1 n (2. whereas in the case of continuous random variables it will take the following form: F (a ) = −∞ ∫ f ( x)dx a (2. Such characteristics summarise and numerically describe certain features for the entire distribution.11 and 2.12.2 Parametric Method In some situations it is easier and even more efficient to look only at certain characteristics of distributions rather than to attempt to specify the distribution as a whole. Probability Theory Figure 2.9.2.15) For the discrete random variables.
then E (cX ) = c × E ( X ) Also. for any two random variables X and Y. The concept of expectation plays an important role not only as a useful measure. For a continuous random variable the expectation is defined as: +∞ E(X ) = −∞ ∫ [1 − F ( x)]dx (2.17) Where the sum is taken for all the values that the variable X can assume. is discrete. however. expected value actually represents the Centre of gravity).20) E ( X + Y ) = E ( X ) + E (Y ) . that these measures serve only to summarise some important features of the probability distribution. (2.19) If c is a constant.2. E(X). they do not completely describe the entire distribution. If the random variable is continuous. In general. One of the most common and useful summary measures of a probability distribution is the expectation of a random variable. It is a unique value that indicates a location for the distribution as a whole (In physical science. It should be remembered. but also as a central concept within the theory of probability and statistics. it is possible to adequately describe a probability distribution with a few measures of this kind. Probability Theory 26 b) Measures of dispersion (or variability) which show the spread of the difference among the possible values of the random variable. then its expectation is defined as: E ( X ) = ∑ x × P( X = x) x (2.18) Where the sum is taken over all values that X can assume. If a random variable. say X. In many cases. the expectation is defined as: E( X ) = +∞ −∞ ∫ x × f ( x)dx (2.
A deviation from the mean.22) The mode. summarises only a certain aspect of a distribution.2. and the probability that X is greater than the median is also 1 2 . Probability Theory Measures of central tendency The most frequently used measures are: 27 The mean of a random variable is simply the expectation of the random variable under consideration. whether it is mode.50 a a +∞ (2. which divides the total area under the PDF into two equal parts. Thus. if P( X ≤ a ) ≥ 0. or any other measure. median. or a listing of possible values of X along with their probabilities is available. mean. the mean value is defined as: Mean = E ( X ) (2. Measures of dispersion The mean is a good indication of the location of a random variable. Thus. A central tendency parameter. It is easy to find two distributions which have the same mean but which are not at all similar in any other respect. expresses the measure of error made by using the mean as a particular value: . D. The median is the point. If a graph of the PMF (PDF). In other words. this can be expressed as: −∞ ∫ f ( x)dx = ∫ f ( x)dx = 0. but no single value need be exactly like the mean.21) The median. X. the probability that X is less than the median is 1 2 . In the continuous case.50 and P( X ≥ a) ≥ 0. is defined as the value of X at which the PDF of X reaches its highest point.50 then 'a' is the median of the distribution of X. is defined as the value of X which is midway (in terms of probability) between the smallest possible value and the largest possible value. for the random variable. determination of the mode is quite simple.
SD = V (X ) (2. x.2. of individual values. More information about them can be found in texts on probability. the more the individual values differ from each other. The solution is to find the square of each deviation from the mean. V.M). and the more apparent the spread within the distribution becomes.24) The positive square root of the variance for a distribution is called the Standard Deviation.23) V ( X ) = E ( X − Mean) 2 = ∫ ( X − Mean) 2 × f ( x )dx if X is continuous (2. Very simply these are expectations of different powers of the random variable. or variability. and then to find the expectation of the squared deviation. The deviation can be taken from other measures of central tendency such as the median or mode. will not work because the expected deviation from the mean must be zero for obvious reasons. X. Consequently. It is quite obvious that the larger such deviations are from a measure of central tendency. The expectation of the deviation about the mean as a measure of variability. . This characteristic is known as a variance of the distribution. thus: V ( X ) = E ( X − Mean) 2 = ∑( X − Mean) 2 × P( x) +∞ −∞ if X is discrete (2. SD. E(X . Probability Theory 28 D= x−M Where. it is necessary to find a measure that will reflect the spread.25) Probability distributions can be analysed in greater depth by introducing other summary measures. known as moments. is a possible value of the random variable.
13.10. In conclusion it could be said that the probability system is wholly abstract and axiomatic.e. In this case the parameter known as coefficient of variation. they do not exist in reality. P(X). CV X .26) Coefficient of variation is very useful because it gives better information regarding the dispersion. because this will depend on the mean value. there are several rules that define the functional relationship between the possible values of random variable X and their probabilities. It is necessary to emphasise that all theoretical distributions represent the family of distributions defined by a common rule through unspecified constants known as parameters of distribution. 2. . As they are purely theoretical. i. The particular member of the family is defined by fixing numerical values for the parameters. Consequently. Instead of analysing the ways in which these rules have been derived. Sometimes it is difficult to use only knowledge of the standard deviation. to decide whether the dispersion is considerably large or small. they are called theoretical probability distributions. which define the distribution.2. every fully defined probability problem has a unique solution. the analysis in this chapter concentrates on their properties. DISCRETE THEORETICAL PROBABILITY DISTRIBUTIONS In probability theory.13 Probability System for Continuous Random Variable Variability The standard deviation is a measure that shows how closely the values of random variables are concentrated around the mean. The concept thus discussed so far is summarised in Figure 2. Probability Theory 29 Figure 2. defined as CV X = SD M (2.
each with some probability. The symbol p. Fortunately.10. then 25 = 32 different sequences of possible outcomes would be observed. Probability Theory 30 The probability distributions most frequently used in reliability. Among the family of theoretical probability distributions that are related to discrete random variables. the probability of the two events. Then the probability of that particular sequence is p 4 q 6 . This is regardless of the order in which successes and failure occur in sequence. If all possible sequences and their probabilities. a car is tested and one of two events. 2. For example. are written down the following fact emerges: The probability of any given sequences of n independent Bernuolli Trials depends only on the number of successes and p. one of the two events is called a “success” and the other a “failure” or “nonsuccess”. and the two event classes and their associated probabilities a Bernuolli Process. each of which can eventuate in only one of two outcomes are known as Bernuolli Trials. exactly 4 success occurs. then the 3 2 1 3 3 4 6 . If 5 independent trials are made (n = 5). the Binomial distribution and the Poisson distribution are relevant to the objectives set by this book. The type of experiment consisting of series of independent trials. since trials are independent. Suppose that in a sequence of 10 trials.2. maintenance and the logistic support are examined in this chapter. it is possible to compute the probability of any sequence. In general. q for the probability of failure (p + q =1). and are not meant to bear any connotation of “goodness” of the event. the probability is p r q n −r where r is the number of successes. stands for the probability of a success. pass or fail. A brief description of each now follows. must occur.1 Bernuolli Trials The simple probability distribution is one with only two event classes. and n − r is the number of failures. These names serve only to tell the events apart. The probability of given sequences depends upon p and q. probability can worked out from: If p = 2 .
28) where: n x n − x n! = p x q n− x p q x x !(n − x )! (2. n. x 0< x <n (2. Generalising this idea for any r. The number of successes.2 The Binomial Distribution The theoretical probability distribution.2. we have the following principle: In sampling from the Bernuolli Process with the probability of a success equal to p. x in n trials is a discrete random variable which can take on only the whole values from 0 through n. This probability distribution is related to experiments. the probability of observing exactly r successes in n independent trials is: n n! P( r successes n.29) The binomial distribution expressed in cumulative form. is called the binominal distribution. By convention the symbol p stands for the probability of a success. The PMF of the Binomial distribution is given by: n PMF ( x ) = P ( X = x ) = p x q n − x . These names are used only to tell the events apart.27) 2.10. Probability Theory 31 The same procedure would be followed for any r successes out of n trials for any p. p) = p r q n −r = p r q n− r r !( n − r )! r (2. which pairs the number of successes in n trials with its probability. each of which can result in only one of two outcomes: success and or failure. and p. which consist of a series of independent trials. representing the probability that X falls at or below a certain value 'a' is defined by the following equation: . q for the probability of failure ( p + q = 1) .
32) .14 PMF and CDF For Binomial Distribution. because of the independence of trials. n = 10.31) Similarly. Figure 2.14 with parameters n = 10 and p = 0. or p(1 − p) summed n times: V ( X ) = np(1 − p) = npq Consequently.30) As an illustration of the binomial distribution.2. the PMF and CDF are shown in Figure 2. the standard deviation is equal to: (2. the variance of the binomial distribution is the sum of the variances of the individual trials. Probability Theory P( X ≤ a ) = 32 ∑ i =o a P( X = xi ) = ∑ i p i q n −i i =0 a n (2.3. p = 0.3 E ( X ) = np (2.
Probability Theory 33 (2.2.5. and negatively skewed if p > 0.5. There are experiments where it is not possible to observe a finite sequence of trials.10. Figure 2. positively skewed if p < 0.5. For example. it is difficult to think of this situation in terms of finite trials. If the number of binomial trials n. is made larger and larger and p smaller and smaller in such a way that np remains constant. the shape of the probability mass function and the cumulative distribution function will depend upon them. 1. Instead. The probability mass function in the case of the Poisson distribution for random variable X can be expressed as follows: e −λ λx P ( X = x λ ) = x! where x = 0.15 shows the PMF of the Poisson distribution with λ = 5 .34) λ is the intensity of the process and represents the expected number of occurrences in a time period of length t.33) Sd ( X ) = npq Although the mathematical rule for the binomial distribution is the same regardless of the particular values which parameters n and p take. 2. say for one minute. then the probability distribution of the number of occurrences of the random variable approaches the Poisson distribution. (2. if the number of cars arriving at a specific junction in a given period of time is observed. 2. observations take place over a continuum. The PMF of the binomial distribution is symmetric if p = 0. such as time.3 The Poisson Distribution The theoretical probability distribution which pairs the number of occurrences of an event in a given time period with its probability is called the Poisson distribution.
the Poisson . Probability Theory 34 Figure 2. the Poisson distribution is a single parameter distribution because it is completely defined by the parameter λ .35) The CDF of the Poisson distribution with λ = 5 is presented in Figure 2.37) Thus. which is equal to λ . In general.2.16. Expected value of the distribution is given by E(X ) = ∑ xP ( X = x ) = x=0 ∑ x x=0 e −λ λx x! Applying some simple mathematical transformations it can be proved that: E(X ) = λ (2.15 PMF of the Poisson Distribution with λ = 5 The Cumulative Distribution Function for the Poisson distribution F ( x ) = P( X ≤ x ) = ∑ i =o x ei λx i! (2.36) which means that the expected number of occurrences in a period of time t is equal to np. The variance of the Poisson distribution is equal to the mean: V (X ) = λ (2.
. Probability Theory 35 distribution is positively skewed.2. although it is nearly symmetrical as λ becomes larger.
2. This means that the Poisson distribution provides a good approximation to the binomial distribution if p is very small and n is large. 2. p → 0 ). the case in which n = 10 and p = 0.2. . the binomial distribution with the parameters n and p. Probability Theory 36 Figure 2. The two distributions agree reasonably well. If more precision is desired. 3. can be approximated to the Poisson distribution with parameter λ = np . p becomes small (that is. The Poisson parameter for the approximation is then λ = np = 10 × 010 = 1 . As an example of the use of the Poisson distribution as an approximation to the binomial distribution. np remains constant. Under these conditions. approximation are shown in Table 2. the Poisson distribution is also a good approximation when p is close to one and n is large. Since p and q can be interchanged by simply interchanging the definitions of success and failure. n becomes large (that is.10 will be considered. The binomial distribution and the Poisson .16 CDF of the Poisson Distribution λ = 5 The Poisson distribution can be derived as a limiting form of the binomial if the following three assumptions were simultaneously satisfied: 1. n → ∞ ). a possible rule of thumb is that the Poisson is a good approximation to the binomial if n / p > 500 (this should give accuracy to at least two decimal places).
which controls the shape of the distribution curves. p = 01) . many distributions may have either one or two parameters): 1.598737 0. Each member of the family is defined with a few parameters. which define the distribution. which defines the origin or the minimum value which random variable. . Shape parameter.074635 0. Probability Theory 37 Table 2.2.606531 0.000965 0. maintainability and supportability engineering are examined in this chapter. CONTINUOUS THEORETICAL PROBABILITY DISTRIBUTIONS It is necessary to emphasise that all theoretical distributions represent the family of distributions defined by a common rule through unspecified constants known as parameters of distribution. Scale parameter. Thus.000158 0 1 2 3 4 5 2.075816 0.012636 0. 2.303265 0. 0. Location parameter also refers to the point on horizontal axis where the distribution is located. individual members of a specific family of the probability distribution are defined by fixing numerical values for the above parameters. The particular member of the family is defined by fixing numerical values for the parameters.2 Poisson Distribution as an Approximation to the Binomial Distribution Binomial P( X = x n = 10.001580 0.11. 3. can have.010475 0. Parameters of a distribution can be classified in the following three categories (note that not all distributions will have all the three parameters. which controls the range of the distribution on the horizontal scale. Each of the above mentioned rules define a family of distribution functions. The probability distributions most frequently used in reliability.315125 0.000061 Poisson P( X = x λ = 1) 0. Source parameter or Location parameter. which in their own way control the distribution.
Figure 2.11.41) . Notice that the exponential distribution is positively skewed. The probability density function of the exponential distribution is given by: f ( x) = λ exp(− λx . x > 0 ) (2.1 Exponential Distribution Exponential distribution is fully defined by a single one parameter that governs the scale of the distribution. Probability density function of exponential distribution for different values of λ The cumulative distribution of exponential distribution is given by: F ( x) = P( X < x) = 1 − exp(− (λx ) ) (2.40) (2.39) It can be shown that the mean and variance of the exponential distribution are: E( X ) = 1/ λ V ( X ) = (1 / λ ) 2 (2.38) In Figure 2.2. with the mode occurring at the smallest possible value. zero.17.17 several graphs are shown of exponential density functions with different values of λ. Probability Theory 38 2.
This can be expressed using the conditional probability as: P{ X > s + t x > t} Using conditional probability of events. 11. the above probability can be written as: P{ X > s + t X > s} = P{ X > s + t ∩ X > t} P{X > s + t} = (2. Assume that we are interested in finding the probability that this item will not fail for another s units of time. the conditional probability depends only on the remaining duration and is independent of the current age of the item. Suppose the present age of the item is t. This property is exploited to a great extend in reliability theory.11. 2.2 Normal Distribution (Gaussian Distribution) This is the most frequently used and most extensively covered theoretical distribution in the literature.42). SD( X ) = E ( X ) = 1 / λ . Suppose that the random variable X measures the duration of time until the occurrence of failure of an item and that it is known that X has an exponential distribution with parameter λ. that is X > t. It has a characteristic symmetrical . Probability Theory 39 The standard deviation in the case of the exponential distribution rule has a numerical value identical to the mean and the scale parameter. The Normal Distribution is continuous for all values of X between − ∞ and + ∞ .1.2. we get P[ X > s + t X > t ] = P[ X > s ] = exp(−λs ) That is.1 Memoryless Property of Exponential Distribution One of the unique property of exponential distribution is that it is the only continuous distribution that has memory less property.42) P{ X > t} P{ X > t} However we know that for exponential distribution P[ X > s + t ] = exp(−λ ( s + t )) and P[ X > t ] = exp(−λt ) Substituting these expressions in equation (2.
43) this becomes: F (a) = ∫ 1 a − µ 2 exp − dx 2 σ − ∞ σ 2π a 1 (2.2.44) . As the deviation of x from the location parameter µ is entered as a squared quantity. The mathematical expression for its probability density function is as follows: 1 x − µ 2 exp − f ( x) = 2 σ σ 2π 1 (2. Probability Theory 40 shape.18. The influence of the parameter µ on the location of the distribution on the horizontal axis is shown in Figure 2. while σ can be any positive finite number. µ and σ also represents the mean and the standard deviation of this distribution. Taking into account Eq. two different x values. This dictates the symmetry of the normal distribution. the median and the mode have the same numerical value. which means that the mean. Parameter µ can be any finite number. will have the same probability density according to this rule. (2. showing the same absolute deviation from µ. where the values for parameter σ are constant.43) Where µ is a location parameter (as it locates the distribution on the horizontal axis) and σ is a scale parameter (as it controls the range of the distribution). The cumulative distribution function for the normal distribution is: F ( a ) = P( X ≤ a ) = −∞ ∫ f ( x )dx a where f(x) is the normal density function.
statisticians have constructed the table of probabilities.19 Cumulative distribution of normal distribution for different values of µ and σ. This is a theoretical random variable with parameters µ = 0 and σ = 1. Z. z= x−µ σ (2.19 several cumulative distribution functions are given of the Normal Distribution. corresponding to different values of µ and σ . (2. As the integral in Eq.2. The relationship between standardised random variable Z and random variable X is established by the following expression: Figure 2.45) . Probability Theory 41 Figure 2.18 Probability density of normal distribution for different σ values In Figure 2. which complies with the normal rule for the standardised random variable.44) cannot be evaluated in a closed form.
for a given z value the table provides the cumulative probability up to.2.48) where Φ is the standard normal distribution Function defined by Φ( z ) = −∞ ∫ x 1 1 exp − z 2 dx 2 2π (2.43) becomes simpler: f (z) = 1 σ 2π 1 − z e 2 2 (2. That is.44) becomes: F (a) = ∫ 1 x−µ exp − z 2 dz = Φ 2 σ − ∞ σ 2π 1 (2. regardless of its particular parameters (see Table in appendix). In Microsoft EXCEL®.49) The corresponding standard normal probability density function is: f ( z) = z2 1 exp − 2 2π (2. and including. (2.46) The standardised form of the distribution makes it possible to use only one table for the determination of PDF for any normal distribution.50) Most tables of the normal distribution give the cumulative probabilities for various standardised values. Probability Theory 42 Making use of the above expression the equation (2. the cumulative distribution .47) By substituting z x−µ σ with z Eq. The relationship between f(x) and f(z) is : f ( x) = f ( z) σ (2. that standardised value in a normal distribution.
20 The areas under a normal distribution between µ .51) V (X ) = σ 2 (2.53) .52) Since normal distribution is a symmetrical about its mean.1 Central Limit Theorem Suppose X1.kσ and µ + kσ 11. σ. Figure 2. and f(x) = NORMDIST (x. − (2.kσ. Probability Theory 43 function and density function of normal distribution with mean µ and standard deviation σ can be found using the following function. σ. the area between µ . X2. µ. Let X − µx Zn = σx /n Where.20. F(x) = NORMDIST (x. which is shown in Figure 2.2.2. FALSE) The expectation of a random variable. µ. … Xn are mutually independent observations on a random variable X having a welldefined mean µx and standard deviation σx. TRUE). is equal to the location parameter µ thus: E( X ) = µ Whereas the variance is (2. µ + kσ (k is any real number) takes a unique value.
The relationship between parameters µ (location parameter of the normal distribution) and µ l is defined: 1 µ = exp µ l + σ l2 2 (2. can in some respects. the probability density function for a random variable X is defined as: f X ( x) = 1 ln x − µ l exp − 2 σ xσ l 2π l 1 2 ≥0 (2.11. be considered as a special case of the normal distribution because of the derivation of its probability function. 2.1). The X values have to be from the same distribution but the remarkable feature is that this distribution does not have to be normal.57) The cumulative distribution function for the lognormal distribution is defined with the following expression: . the random variable X follows the lognormal distribution. If a random variable Y = ln X is normally distributed then.3 Lognormal Distribution The lognormal probability distribution. Thus. gamma.56) The parameter µ l is called the scale parameter (see Figure 2.55) where FZ (z) is the cumulative distribution of standard normal distribution N(0.21) and parameter σ l is called the shape parameter. Weibull or even an unknown one. exponential. . beta. n→∞ lim FZ n ( z ) = FZ ( z ) (2.∞ < z < ∞. Then for all z.2. Probability Theory 44 X = − 1 n ∑ Xi n i =1 (2.54) and Fz n (z ) be the cumulative distribution function of the random variable Zn. it can be uniform.
Then. Probability Theory 45 Figure 2. making use of the standardised random variable Equation (2.21 Probability density of lognormal distribution FX ( x) = P( X ≤ x) = ∫ 1 ln x − µ l exp − 2 σ l 0 xσ l 2π x 1 2 dx (2.2.61) transforms into: ln x − µ l FX ( x) = P( X ≤ x) = Φ σ l (2.59) The measures of central tendency in the case of lognormal distributions are defined by the: (a) Location parameter (Mean) 1 M = E ( X ) = exp µ l + σ l2 2 (b) Deviation parameter (the variance) (2. the same procedure is applied as in the case of normal distribution.58) As the integral cannot be evaluated in close form.60) .
The shape of this distribution is governed by its parameter. γ > 0. fibre strength of Indian cotton. it has a very important role in the statistical analysis of experimental data. As the Weibull distribution has no characteristic shape. The rule for the probability density function of the Weibull distribution is: β x −γ f ( x) = η η β −1 x −γ exp − η β (2.4 the Weibull approximates to the normal distribution. when β = 3.11.Weibull. in such cases: f ( x) = β η x η β −1 x β exp − η (2. β. As the location parameter ν is often set equal to zero. and the fatigue life of a St37 steel by the Swedish engineer W. For example.66) By altering the shape parameter β. it is identical to the exponential distribution. Figure 2. the Weibull distribution takes different shapes.61) 2. the size distribution of fly ash. Probability Theory 46 2 V ( X ) = exp 2 µ l + σ l ( [exp(σ ) − 1) 2 l ] (2.4 Weibull Distribution This distribution originated from the experimentally observed variations in the yield strength of Bofors steel.65) where η. The cumulative distribution functions for the Weibull distribution is: . such as the normal distribution.22 shows the Weibull probability density function for selected parameter values. when β =1.2.
1.22. η = 0. Probability density of Weibull distribution with β = 2.0. 2 For γ = 0. Probability Theory 47 x−γ F ( x) = 1 − exp − η B (2.67) Figure 2.2. γ = 0.68) 1 E ( X ) = γ + η × Γ + 1 β where Γ is the gamma function. the cumulative distribution is given by x β F ( x ) = 1 − exp − η The expected value of the Weibull distribution is given by: (2.69) .5. defined as (2.
For other values. The variance of the Weibull distribution is given by: 2 1 V ( X ) = (η ) 2 Γ1 + − Γ 2 1 + β β (2. one has to solve the above integral to the value. In Microsoft EXCEL. Probability Theory 48 Γ ( n) = ∫ e − x × x n −1dx 0 ∞ When n is integer then Γ (n) = (n − 1)!. Γ (x) can be found using the function. Gamma function.2. EXP[GAMMALN(x)].70) . Values for this can be found in Gamma function table given in the appendix.
Mission Reliability Measures are used to predict the system's ability to complete mission. Basic Reliability Measures. Use of any particular reliability measure depends on what is expected of the system and what we are trying measure. 2. Several life cycle decision are made using reliability measure as one of the important design parameter. environment.3. failure free operating period (FFOP). Reliability measures like reliability function and failure function fall under this category. Operational Reliability Measures are used to predict the performance of the system when operated in a planned environment including the combined effect of design. These measures consider only those failures that cause mission failure. etc. Operational Reliability Measures.1) as: 1. Requirements must be tailored to individual item considering operational environment and mission criticality. . Contractual Reliability Measures. the reliability metrics can be classified (Figure 3. and it works Lincoln Steffens In this chapter we discuss various measures by which hardware and software reliability characteristics can be numerically defined and described. support policy. Mission Reliability Measures. quality. Basic Reliability Measures are used to predict the system's ability to operate without maintenance and logistic support. In broader sense. Reliability Measures 49 Chapter 3 Reliability Measures I have seen the future. and 4. maintenance free operating period (MFOP). The reliability characteristics or measures used to specify reliability must reflect the operational requirements of the item. Reliability measures such as mission reliability. maintenance. and hazard function fall under this category. 3. Manufacturers and customers use reliability measure to quantify the effectiveness of the system.
Currently. Maintenance Free Operating Period (MFOP). Measures such as Mean Time To Failure (MTTF). However. All the measures are defined based on the assumption that the timetofailure (TTF) distribution of the system is known. Mean Time Between Overhaul (MTBO). Recent projects such as Future Offensive Air Systems (FOAS) drive maintenance free operating periods (MFOP) as the preferred reliability requirement. Basic Reliability Contractual Reliability Reliability Measures Operational Reliability Mission Reliability Figure 3. Mean Time Between Failure (MTBF) and Failure rate fall under this category. Selection of specific measure to quantify the reliability requirements should include mission and logistic reliability along with maintenance and support measures. Contractual reliability is calculated by considering design and manufacturing characteristics. . In the next Section. Contractual Reliability Measure is used to define. one may require more than one reliability metric in most of the cases for specifying reliability requirements. Basically it is the inherent reliability characteristic. Procedures for finding the timetofailure distribution by analysing the failure data that are discussed in Chapter 12. MTBF and failure rates have several drawbacks. we define various reliability measures and how to evaluate them in practical problems. Mean Time Between Critical Failure (MTBCF) and Mean Time Between Unscheduled Removal (MTBUR) fall under this category. many manufacturers specify reliability by using mean time between failure (MTBF) and failure rate. measure and evaluate the manufacturer's program.3.1 Classifications of Reliability Measures Though we classify the reliability measures into four categories as mentioned above. Reliability Measures 50 Measures such as Mean Time Between Maintenance (MTBM).
number of cycles. Weibull. equation (3. and gives the probability of failure free operation without maintenance up to time t. F(t) = P (failure will occur before or at time t) = P (TTF ≤ t) = ∫ f (u )du 0 t (3.2a. Since negative time is meaningless in reliability. see Figure 3. The failure function is usually represented as F(t). most of the complex systems will require maintenance at frequent intervals. Exponential.12. Failure function is equal to the probability that the timetofailure random variable will be less than or equal a particular value t (in this case operating time. normal. Reliability Measures 51 3. For µ >> 3σ. etc. great care should be taken in using normal distribution for the failure function. Gamma and Gumbel are few popular theoretical distributions that are used to represent failure function.1) f (t ) Failure density F(t) Figure 3.2a).1) is derived by assuming that no maintenance is performed to the system.3. Failure functions of few popular theoretical distributions are listed in Table 3. FAILURE FUNCTION Failure function is a basic (logistic) reliability measure and is defined as the probability that an item will fail before or at the moment of operating time t. In such cases. Here time t is used in a generic sense and it can have units such as miles. Failure function of a hypothetical distribution Where f (t ) is the probability density function of the timetofailure random variable TTF. so may have significant value at t ≤ 0. to incorporate the behaviour of the system under maintenance. It should be noted that in case of normal distribution the failure function exists between ∞ and +∞. However. depending on the operational profile and the utilisation of the system. Equation (3. flying hours..1) has to be modified. lognormal. Time .1. That is. number of landings. probability values for t ≤ 0 can be considered negligible.
TRUE) in EXCEL® Weibull 1 − exp( − ( t −γ β ) ) η η. F(0) = 0. λ > 0 Normal 0σ ∫ t 1 2π 1 x− µ −[ ] e 2 σ dx 2 or t − µ Φ σ or NORMDIST(t. t ≥ γ Gamma 1 β α x α −1e −βx dx Γ(α ) ∫ 0 t Note that the failure function of normal distribution is defined between 0 and t. However. µ. F (t1 ) ≤ F (t2).3. That is. of few theoretical distributions Distribution Failure Function. µ.3c.2b). Failure function is an increasing function.3a3. σ. . For modelling purposes it is assumed that the failure function value at time t = 0. 2. Reliability Measures Table 3.γ > 0 . F(t). for t1<t2. since t is greater than 0 for reliability purposes (against the usual limit ∞) Applications of failure function are listed below (Figure 3. Characteristics of failure function 1. Failure functions of various theoretical distributions for different parameter values are shown in Figures 3. this assumption may not be valid always.1 Failure function. σ. β . TRUE) in EXCEL® Lognormal 1 ln( x ) − µ l − 2 σ l e 2 0σl x ∫ t 1 2π dx or ln(t ) − µ l Φ σ l or NORMDIST(ln(t). F(t) 52 Exponential 1 − exp( −λt ) t > 0.
Failure Function Increasing function Probability of failure by given age Fraction of items that fail by given age Figure 3. The value of failure function increases as the time increases and for t = ∞. Applications of failure function 1. systems can be dead on arrival.2 0 Figure 3.F(t) is the probability that an individual item will survive up to time t.3. F(t) is the probability that an individual item will fail by time t. 1 .4 0.6 λ = 0. 3.8 0.2b.03 λ = 0. F(t) 1 0. Reliability Measures 53 For example.01 0 100 200 Time 300 400 0. F(t) is the fraction of items that fail by time t. 2.02 λ = 0. Properties of failure function Failure Function. F(∞) = 1.3a: Failure function of exponential distribution for different values of λ .
3.05. F(t) 1 µ = 100 µ = 120 µ = 140 0. b) Find the maximum length of flight such that the failure probability is less than 0.8 0.1 The time to failure distribution of a subsystem in an aircraft engine follows Weibull distribution with scale parameter η = 1100 flight hours and the shape parameter β = 3. . Find: a) Probability of failure during first 100 flight hours. Reliability Measures 54 Failure Function.3c Failure function of normal distribution for different µ values Example 3.4 0 0 0. F(t) β=1 β=2 β=3 Time Figure 3.3b Failure function of Weibull distribution for different β values Failure Function.6 0.2 50 100 Time 150 200 Figure 3.
we get t = 408. In this case.05 F (t ) = 1 − exp(−( 1100 t 3 ) ) > 0.2 The time to failure distribution of a Radar Warning Receiver (RWR) system in a fighter aircraft follows Weibull distribution with scale parameter 1200 flight hours and shape parameter 3. we have t −0 3 ) ) < 0. Reliability Measures SOLUTION: a) The failure function for Weibull distribution is given by: F (t ) = 1 − exp( − ( t −γ β ) ) η 55 It is given that: t = 100 flight hours. If the supplier gives a warranty for 750 flight hours. β = 3 and γ = 0.95 ⇒ t = 1100 × [− ln(0.001.95)]1 / 3 1100 Now solving for t. (Although we have a same system.95 = exp(−( 1100 t 3 =( ) > − ln 0. calculate the risk involved with respect to fighter aircraft and the helicopter.00075 1100 b) If t is the maximum length of flight such that the failure probability is less than 0.05 is 408.70 flight hours.70 flight hours. RWR in helicopter is subject to more vibrations compared to aircraft). The time to failure distribution of the same RWR in a helicopter follows exponential distribution with scale parameter 0.3. The maximum length of flight such that the failure probability is less than 0. Probability of failure within first 100 hours is given by: F (100) = 1 − exp( − ( 100 − 0 3 ) ) = 0. the operating conditions have significant impact on the failure function.05. η = 1100 flight hours. Example 3. Compare the failure function of the RWR in the fighter aircraft and the helicopter. .
3.4 depicts the failure function of RWR in fighter aircraft and the helicopter.001 × t )) Figure 3. . Reliability Measures SOLUTION: The failure function of RWR on the fighter aircraft is given by: F (t ) = 1 − exp( − ( t 3 ) ) 1200 56 The failure function of RWR on the helicopter is given by: F (t ) = 1 − exp( −( 0.
more than 52% of the RWR’s are likely to fail before the warranty period.3.6 Helicopter Aircraft 0. Reliability Measures 57 1 0.2166 1200 That is. If the RWR is installed in helicopter then the associated risk is given by: F (750) = 1 − exp( −0.12. just above 21% percent of RWR are likely to fail if the RWR is installed in the aircraft.5276 In the case of helicopter.8 Failure Function.4 Failure function of RWR in fighter aircraft and helicopter If the supplier provides warranty for 750 flight hours the risk associated with aircraft is given by: F (750) = 1 − exp( − ( 750 3 ) ) = 0. In most of the cases there will be more than one (some times .4 0.001 × 750) = 0. 3.2 0 0 1000 2000 Time 3000 4000 Figure 3.1 Failure function of system under multiple failure mechanisms It is seldom true that an item's failure is caused by a single failure mechanism. F(t) 0.
005 hours. The expression (3. 1 − FB (t ) = exp( − λ B t ) Now the failure function of the item is given by: . F1(t) and F2(t) the are failure function for failure mechanism 1 and 2 respectively.002 hours. most of the practical systems fail due to different causes or different failure mechanisms. 1 − F A (t ) = exp( − λ A t ) f B ( t ) = λ B exp( − λ B t ) . Find the probability that the item will fail before 500 hours of operation. Reliability Measures 58 hundreds) mechanism that causes the failure of an item. SOLUTION: Assume that fA(t) and fB(t) represent probability density function of the timetofailure random variable due to failure mechanism A and B respectively.1) is more appropriate when the failure is caused by a single failure mechanism. Now the probability density function of the timetofailure of the system caused by either of the failure mechanisms: f (t ) = f1 (t )[1 − F2 (t )] + f 2 (t )[1 − F1 (t )] where. f A ( t ) = λ A exp( − λ A t ) . Thus.2) Example 3. The timetofailure distribution of the item due to failure mechanism B can be represented by exponential distribution with parameter λB = 0. The failure function of the item under two different failure mechanism is given by: F (t ) = ∫ { f1 ( x)[1 − F2 ( x)] + f 2 ( x)[1 − F1 ( x)]}dx 0 t (3.3. Let f1(t) and f2(t) be the probability density function of the system due to failure mechanism 1 and 2 respectively. Assume that the system failure is due to two different failure mechanisms. The timetofailure distribution of the item due to failure mechanism A can be represented by exponential distribution with parameter λA = 0.3 Failure of an item is caused by two different failure mechanisms (say failure mechanism A and B). However.
The probability that the item will fail by 500 hours is given by: F (500) = 1 − exp(−((0. t]} = 1 .5 represents the failure function due to failure mechanism 1. F(t) 0.2 0 0 1000 Failure mechanism A Failure mechanism B System 2000 3000 Time Figure 3. t.3) .6 0.3.8 0. RELIABILITY FUNCTION Reliability is the ability of the item to maintain the required function for a specified period of time (or mission time) under given operating conditions. Reliability Measures 59 F (t ) = ∫ {λ A exp(−(λ A + λ B ) x) + λ B exp( −(λ A + λ B ) x)dx 0 t = (λ A λ A + λ B )[1 − exp(−(λ A + λ B )t ] + (λ B λ A + λ B )[1 − exp(−( λ A + λ B )t ] = [1 − exp( −(λ A + λ B )t ] Figure 3.4 0. is defined as the probability that the system will not fail during the stated period of time.13. 2 and the system failure function.5 Failure function due to different failure mechanisms 3.9698 1 Failure Function. R(t). If TTF represents the timetofailure random variable with failure function (cumulative distribution function) F(t).F(t) (3.005 + 0. Reliability function. then the reliability function R(t) is given by: R(t) = P{the system doesn't fail during [0 .002) × 500)) = 0. under stated operating conditions.
7ac represents reliability function of various theoretical distributions for different parameter values. . As t becomes larger and larger R(t) approaches zero. that is. 2.6 Reliability function of a hypothetical probability distribution Properties of reliability function: 1. then we have to calculate mission reliability function. the timetofailure follows exponential distribution due to memory less property of exponential distribution). R(∞). R(t) is the area under TTF density between t and ∞.3) is valid only for new systems or those systems whose failures are not age related (that is. which will be discussed later. Figure 3. R(t) is the fraction of items in a population that survive up to time t. Reliability Measures 60 In equation (3. R(t) is the basic function used for many reliability measures and system reliability prediction. If the system age is greater than zero at the beginning of the mission. in most of the cases this assumption may not be valid.3. 2.6 depicts the relation between reliability function and the TTF density function. Reliability function for some important life distributions are given in Table 3. Applications of reliability function 1. R(t1) ≥ R(t2). However. That is. f (t ) TTF Failure density R(t) Time Figure 3. 3. Reliability is a decreasing function with time t. Figure 3. Thus the equation (3. It is usually assumed that R (0) = 1.2. R(t) is the probability that an individual item survives up to time t. for t1 < t2 .3) we assume that the age of the system before the start of the mission is zero.
t. Reliability function. Reliability Measures 61 Table 3. t ≥ γ Gamma 1− 1 t α α −1 − βx e dx ∫β x Γ(α ) 0 Reliability Function. σ. β . R(t) 1 0. λ > 0 Normal Φ( µ −t 1 ) =1− ∫ σ 0 σ 2π t 1 x − µ 2 − 2 σ dx e or NORMDIST (µ. σ.001 λ λ=0. TRUE) in EXCEL Weibull exp(−( t −γ β ) ) η η.2 0 λ=0. γ > 0 .2.008 0 1000 Time 2000 3000 . R(t).005 = λ=0. for popular theoretical distributions Distribution Reliability function.6 0. TRUE) in EXCEL 1 ln( x ) − µ l − 2 σl e 2 Lognormal Φ( µ l − ln t 1 ) =1− ∫ σl 0 σ l x 2π t dx or NORMDIST (µ. ln(t).4 0.8 0.3. R(t) Exponential exp(−λt ) t > 0.
Reliability Measures 62 Figure 3.7 a. Reliability function of exponential distribution for different values of λ .3.
6 0.6 β =3 β =2 β =1 0. SOLUTION Using Table 3.3.7c.5) = 0.2.7 b. Find the reliability of this chip for a mission of 8000 hours. R(t) µ=50 µ=60 µ=70 R(t ) = Φ( µ−t 9000 − 8000 ) = Φ( ) = Φ(0.2 0 0 50 Time Figure 3.2 0 0 50 100 150 Time 200 250 Figure 3.8 0. Reliability function of Normal distribution for different values of µ Example 3.6915 σ 2000 . Reliability function of Weibull distribution for different values of β 1 0.4 0. the reliability for a mission of 8000 hours is given by: 100 150 Reliability Function.4 Time to failure distribution of a computer memory chip follows normal distribution with mean 9000 hours and standard deviation 2000 hours. Reliability Measures 63 Reliability Function.8 0.4 0. R(t) 1 0.
Let f1(t) and f2(t) be the probability density function of the timetofailure random variable due to failure mechanism 1 and 2 respectively. Reliability Measures Example 3.2.1.3.4) 0 t The above result can be extended to obtain expression for reliability function due to more than two failure mechanisms. find the reliability of the item for 200 hours.3.1 ) = 0.13.5 64 The time to failure distribution of a steam turbo generator can be represented using Weibull distribution with η = 500 hours and β = 2.1 Reliability function for items under multiple failure mechanisms Assume that the failure of the item is caused due to two different failure mechanisms. The Reliability function of the item under two different failure mechanism is given by: R (t ) = 1 − F (t ) = 1 − ∫ { f1 ( x)[1 − F2 ( x)] + f 2 ( x)[1 − F1 ( x)]}dx (3. The probability density function of the timetofailure of the item is given by caused by either of the failure mechanisms: f (t ) = f1 (t )[1 − F2 (t )] + f 2 (t ) ⋅ [1 − F1 (t )] Where F1(t) and F2(t) are failure function for failure mechanism 1 and 2 respectively.6 For the example 3.2307 3. SOLUTION: Again using Table 3. Example 3. Find the reliability of the generator for 600 hours of operation. reliability of the generator for 600 hours of operations is given by: R( t ) = exp( − ( 600 / 500) 2. SOLUTION: .
The main difference between reliability function and the mission reliability function is that. MR(ta.3. If the time to failure distribution is exponential. t m ) = R (t m ) Application of mission reliability function 1. the reliability function can be written as: R(t ) = exp(−(λ A + λ B ) × t ) R(200) = exp(−(0.002 + 0. in mission reliability we recognise the age of the system before the mission. The expression for mission reliability MR (tb . then the following relation is valid.13. Example 3.7 .2465 3. We assume that no maintenance is performed during the mission. Reliability Measures 65 Using the expression for failure function obtained in example 3.005) × 200) = 0. Mission reliability is defined. tm) is given by MR(t b . one might be interested in finding the probability of completing a mission successfully.3. Success probability of hitting an enemy target and returning to the base is an example where mission reliability function can be used. t b is the age of the item at the beginning of the mission and tm is the mission period. as the probability that the system aged tb is able to complete mission duration of tm successfully.5) where. MR (t b .2 Mission Reliability Function In many practical situations. tm) gives the probability that an individual item aged ta will complete a mission duration of tm hours without any need for maintenance. Mission reliability is the appropriate basic reliability measure for ageing items or items whose timetofailure distribution is other then exponential. Mission reliability. 2. t m ) = R (t b + t m ) R(t b ) (3.
25.200) = R(1700) 0.3. it is estimated that delay cost per minute for . 3. For most airlines.5221 2400 R(1700) = exp(−( R(1500) = exp(−( 1500 1. Find the probability that that gearbox will not fail during a mission time of 200 miles. SOLUTION: Given. Airlines frequently seek DR guarantees where the aircraft manufactures face penalties if DR levels are not achieved. Assuming that the age of the gearbox is 1500 miles.5736 2400 MR(1500.5736 That is.5221 = = 0. For commercial airlines despatch reliability is an important economic factor. tb = 1500 miles and tm = 200 miles MR(t b . DESPATCH RELIABILITY Despatch reliability (DR) is one of popular reliability metrics used by commercial airlines around the world.14. Despatch reliability is defined as the percentage of revenue departures that do not occur in a delay or cancellation due to technical problems.9102 R(1500) 0. Reliability Measures 66 Timetofailure distribution of the gearbox within an armoured vehicle can be modelled using Weibull distribution with scale parameter η = 2400 miles and shape parameter β = 1. the gearbox aged 1500 miles has approximately 91% chance of surviving a mission of 200 miles.25 ) ) = 0. the delay means that the aircraft is delayed more than 15 minutes.25 ) ) = 0. Technical delays occur can be caused due to some unscheduled maintenance. t m ) = R(t m + t b ) R (1700) = R (t b ) R (1500) 1700 1.
it is the limiting value of the probability.15. h(t)δt. is not a probability. it represents the conditional probability of failure in an interval t to t + δt given that the system survives up to t. It quantifies the risk of failure as the age of the system increases. Hazard function is the indicator of the effect of ageing on the reliability of the system.6) is applied only to technical delays. system and component maintainability. δtR(t ) R(t ) δt → 0 δt δt → 0 (3. and overall logistic support. h(t ) = lim R(t ) − R(t + δt ) 1 F (t + δt ) − F (t ) = lim . represents the probability that the item will fail between ages t and t+δt as δt →0.7) Note that hazard function. However. The expression for despatch reliability is given by: DR(%) = Where. The above expression can be simplified so that h(t ) = f (t ) R(t ) (3.3.6) ND15 = Number of delays with more than 15 minutes delay NC = the number of cancellations Equation (3. Mathematically. h(t).8) . 3. Reliability Measures 67 large jets can be as high as 1000 US dollars. as δt tends to zero. divided by δt. 100 − ND15 − NC × 100% 100 (3. DR is a function of equipment reliability. HAZARD FUNCTION (HAZARD RATE OR INSTANTANEOUS FAILURE RATE) Hazard function (or hazard rate) is used as a parameter for comparison of two different designs in reliability theory. that is.
3. is valid for all types of time to failure distribution. h(t). decreasing or constant. Hazard function shows how the risk of the item in use changes over time (hence also called risk rate). it immediately follows that: (3.3. Hazard function is not a probability and hence can be greater than 1. The hazard functions of some important theoretical distributions are given in Table 3.10). the hazard function is the ratio of the probability density function to the reliability function. 2.3.9). Characteristics of hazard function 1. of few theoretical distributions Distribution Hazard function.9) f (t ) = h(t ) exp( − ∫ h ( x )dx 0 t (3.10) The expression (3. Hazard function. Hazard function can be increasing. which relates reliability and hazard function. we get: t ∫ h( x )dx = ∫ t f ( x) 0 = ∫− 0 0 R( x ) t R '( x ) dx dx = − ln R (t ) R( x) Thus reliability can be written as: t R(t ) = exp − ∫ h ( x)dx 0 From equation (3. Table 3. Integrating both sides of the above equation. Reliability Measures 68 Thus. h(t) λ Exponential .
0 2 0 0 50 100 Tim e 150 200 250 hazard function β = 2. h(t) 69 Normal f (t ) / Φ ( µ −t ). fl(t) is the pdf of lognormal Weibull β t β −1 ( ) η η [ 1 t α α − 1 − βx β α α − 1 − βt t e ] /1− e dx ∫β x Γ (α ) Γ (α ) 0 Gamma Applications of hazard function 1. σ Lognormal µ −t f l (t ) / Φ( l ). 3.4 β = 0.0 4 0 . f(t) is the pdf of normal distribution.6 . For h(t) ≤ 1.8 β = 1. 0 . σl distribution. Figures 3.0 8 0 . it is not recommended to carry out preventive maintenance. h(t) is loosely considered as failure rate at time t (timedependent) 2. Reliability Measures Distribution Hazard function.0 6 0 .8ac show hazard function of various theoretical distributions for different parameter values.3. h(t) quantifies the amount of risk a system is under at time t.
SOLUTION: The hazard rate for Weibull distribution is given by: .8a Hazard function of Weibull distribution for different values of β 0. Reliability Measures 70 Figure 3.02 hazard function 0. Find the hazard rate of the gas turbine at time t = 800 hours and t = 1200 hours.8b Hazard function of exponential distribution 0.2 0.3.4 0.015 0.8 Time to failure distribution of a gas turbine system can be represented using Weibull distribution with scale parameter η = 1000 hours and shape parameter β = 1.3 0.005 0 0 50 100 Tim e 150 200 Figure 3.1 0 0 20 40 T im e hazard function µ = 50 µ = 60 µ = 70 60 80 100 Figure 3.6 0.5 0.01 0.7.8c Hazard function of normal distribution for different values of µ Example 3.
2 Cumulative hazard function and the expected number of failures Consider an item. Reliability Measures h (t ) = β t β −1 ( ) η η 71 h(800 ) = 17 800 0. which upon failure is subject to minimal repair. Cumulative hazard function.1 Cumulative hazard function Cumulative hazard function represents the cumulative hazard or risk of the item during the interval [0. It can be shown that under the assumption that the item receives minimal repair* (‘asbad asold’) after each failure.13) * Mathematically minimal repair or as bad as old means that the hazard rate of the item after repair will be same as the hazard rate just prior to failure.7 ( ) = 0.11) Reliability of an item can be conveniently written using cumulative hazard as: R (t ) = e − H ( t ) (3.7 1200 0. then E[N(t)] = M (t ) = ∫ h( x) dx 0 t (3.0019 1000 1000 h(1200 ) = 3.12) 3.t]. That is.7 . then M(t) = E [N(t)] is the expected number of failures by time t. . is given by: H (t ) = h( x ) dx 0 ∫ t (3. If N(t) is the total number of failures by time t. ( ) = 0. H(t).15.3.15. the hazard rate after repair is same as the hazard rate just before failure.00145 1000 1000 1.
9 An item is subject to minimal repair whenever it failed. Reliability Measures 72 The above expression can be used to model different maintenance/replacement policies.14) Weibull time to failure distribution E[N(t)] = ∫ 0 t h( x )dx = ∫ η (η ) 0 t β x β −1 t dx = ( ) β η (3. then the expected number of failures is given by the renewal function. If the cost of minimal repair is $ 100 per each repair. The number of times the item is expected to fail by 1500 hours. Find: 1. The expected number of failures is given by: 1500 2 t ] = 32 = 9 E[N(t)] = [ ] β = [ 500 η 2. The cost of the item is $ 200. If the time to failure of the item follows Weibull distribution with η = 500 and β = 2. where . M(t) [refer chapter 4]. In case of exponential and Weibull time to failure distributions we get the following simple expressions for the expected number of failures of an item subject to minimal repair. the expected number of failures is given by E[N(t)] = ∫ h( x ) dx = ∫ λdx = λt 0 0 t t (3. Crepair (t) = 9 × 100 = $ 900.3.15) Example 3. Exponential time to failure distribution For exponential distribution. SOLUTION: 1. and 2. Using the above result the cost associated with repair. is it advisable to repair or replace the item upon failure. If the item is replaced.
15. Early operation will remove these items from a population of like items. However. The constant or gradually increasing region is followed by wear out region characterised by increasing hazard function. Pattern A is called the bathtub curve and consist of three distinct phases. Pattern E represents a slowly increasing hazard without any sign of . The remaining items have a constant hazard for some extended period of time during which the failure cause is not readily apparent. Early failure region is followed by constant or gradually increasing region (called useful life). It was believed that bathtub curve represents the most general form of the hazard function. the recent research shows that in most of the cases hazard function do not follow this pattern. One would expect bathtub curve at the system level and not at the part or component level (unless the component has many failure modes which have different TTF distribution). 3. Finally those items remaining reach a wearout stage with an increasing hazard rate. Thus. It starts with early failure region (known as burnin or infant mortality) characterised by decreasing hazard function. Figure 3. Pattern D shows constant hazard throughout the file.3 Typical Forms of Hazard Function In practice. Pattern C starts with a constant or slowly increasing failure probability followed by wear out (sharply increasing) hazard function. Depending on the equipment and its failure mechanism. Recent research in the field of reliability centred maintenance (RCM) shows that the hazard rate mostly follows six different patterns.3. one can say that the hazard function may follow any one of these six patterns. one should not blindly assume that hazard rate of any item will follow any one of these six patterns. Reliability Measures M (t ) = 73 ∑ F i (t ) i =1 ∞ For the above case. hazard function can have different shapes. it is better to replace the item upon failure rather using minimal repair.9 shows most general forms of hazard function. The reason for such as shape is that the early decreasing hazard rate results from manufacturing defects. However. Thus the cost due to replacement will be less than 4 × 200 = $ 800. the value of M(t) < 4 (The actual calculation of the above function will be discussed in Chapter 4). These are only possible cases based on some data. Pattern B starts with high infant mortality and then follows a constant or very slowly increasing hazard function.
Reliability Measures 74 wear out.01 0 0. Relationship between F(t).02 0.01 0 0 20 40 Age 60 80 100 0 100 Age 200 300 Pattern C: Slowly increasing hazard followed by wearout 0.4 shows the relationship between failure function. Pattern A: Hazard function (bathtub curve) Pattern B: High infant Mortality 0. Different forms of hazard function Table 3.004 0. R(t) and h(t) F(t) R(t) h(t) F(t)  1 .008 0.0006 0.015 0.0101 0.3. Pattern F starts with a low hazard initially followed by a constant hazard.R(t) 1 − exp( − ∫ h ( x) dx) 0 t .0004 0.03 0.00995 0 50 Age 100 150 Pattern E: Slowly increasing hazard function 0.01 0.02 0.03 0. reliability function and hazard function. Table 3.01005 0.04 0.0002 0 0 50 Age 100 150 Pattern D: Constant Hazard 0.9.01 0.005 0 0 50 100 150 200 250 Age Figure 3.4.01 0.012 0.006 0.002 0 0 50 Age 100 150 Pattern F: Initial low hazard followed by constant hazard 0.
for good estimation one has to observe the system failure for a sufficiently large operating period.16) Care should be taken in using the above equation. Failure rate = Total number of failures in a sample Cumulative operating time of the sample (3.t] is λt. Many defence standards such as MILHDBK217 and British DEFSTAN 0040 recommend the following equation for estimating the failure rate.15. . then the expected number of items that fail in [0. 2. 3. Also. failure rate is a special case of hazard function (which is time dependent failure rate). it is appropriate to use failure rate only when the timetofailure distribution is exponential.F(t)  exp( − ∫ h( x)dx 0 t h(t) F ' (t ) [1 − R(t )] − R ' (t ) R(t )  3. That is. By definition. failure rate can be used only for a nonrepairable system. If the failure rate is λ. Failure rate is one of the most widely used contractual reliability measures in the defence and aerospace industry. we call it as failure rate.3. Applications of failure rate 1. Reliability Measures F(t) R(t) h(t) 75 R(t) 1 . Failure rate is one of the popular contractual reliability measures among many industries including aerospace and defence.4 Failure rate Whenever the hazard function is constant. Failure rate represents the number of failures per unit time.
'We know exactly how many tanks to send. What many people do not realise is that MTTF is only a measure of central tendency.2 0 1050 1200 1350 1500 1650 1800 1950 2100 2250 150 300 450 600 750 900 0 Time Figure 3. Unfortunately there are many misconception about MTTF among reliability analysts. One may have to use numerical approximation such as trapezium approach to find MTTF value.10 depicts the MTTF value of an item.17) Thus. MTTF can be defined as: ∞ MTTF = tf ( t ) dt = 0 ∫ ∞ ∫ R(t )dt 0 (3.4 MTTF 0.3. between zero and infinity. one of Generals from a defence department said.16.8 0.6 R(t) 0. we measured the distance from the map and divided that by MTTF'. During the Gulf War. Figure 3. MTTF can be considered as the area under the curve represented by the reliability function. . If the item under consideration is repairable. It is used as a measure of reliability for nonrepairable items such as bulb. it is difficult to evaluate the integral (3.17) represents mean time to first failure of the item. R(t). 1 0. then the expression (3.10 MTTF of an item as a function of Reliability MTTF is one of the most popular measures for specifying reliability of nonrepairable items among military and Government organisations throughout the world.17). microchips and many electronic circuits. MEAN TIME TO FAILURE (MTTF) MTTF represents the expected value of a system's time to first failure. Mathematically. Reliability Measures 76 3. For many reliability functions.
11 Comparison of item with same MTTF Using the equation (3. if the timetofailure distribution is exponential. For a repairable system.4 0. One might think that both the components have equal reliability.2 0 0 200 400 600 800 1000 Exponential Normal Time Figure 3. MTTF is the average life of a nonrepairable system.5. MTTF is one of the important contractual reliability measures for nonrepairable (consumable) items. then 63% of the items will fail before their age reaches MTTF value.6 0. The figure clearly shows that items with exponential failure time show higher chance of failure during the initial stages of operation. it will be 50%. Figure 3. For example let us assume that we have two items A and B with same MTTF (say 500 days).8 0. It is easy to check that if the time to failure of the item is exponential then more than 63% of the items will fail by the time the age of the item reaches MTTF. MTTF represents the average time before the first failure. However. Failure Function 1 0. Reliability Measures 77 For example. However. it is important to understand what MTTF value really means.11 shows the cumulative distribution of these two items up to 500 days.3. the MTTF of various failure distributions are listed in Table 3. if the time to failure of the item A is exponential is that of item B is normal then there will be a significant variation in the behaviour of these items. .17). Applications of MTTF 1. In the case of normal distribution. 2.
MTTF(t0) can be expressed as: ∞ MTTF (t 0 ) = (t − t 0 ) f ( t  t 0 )dt t0 ∫ (3. provided that the item has survived over time t0. MTTF is one of the popular contractual reliability measures for nonrepairable systems. Mathematically. f (t  t 0 ) = h (t ) × R (t  t 0 ) . Thus. MTTF of different timetofailure distributions Distribution Exponential Normal MTTF 1/λ µ 78 Lognormal σ exp(µ l + l ) 2 Weibull η × Γ(1 + α/β 1 ) β Gamma 3. which represents the expected time to failure of an item aged t0. Reliability Measures 3. mean residual life).3.18) f(tt0) is the density of the conditional probability of failure at time t.1 Mean Residual Life In some cases. Table 3.5.16. it may be of interest to know the expected value of the remaining life of the item before it fails from an arbitrary time t0 (known as. We denote this value as MTTF(t0).
10 Companies A and B manufacture car tyres.0005 and the time to failure distribution of B is normal with µ = 2000 miles and σ = 200 miles.20) The concept of mean residual life can be successfully applied for planning maintenance and inspection activities. Now. Both the companies claim that the MTTF of their car tyre is 2000 miles. given that it has survived up to time t0.19) substituting for h(t) in the above equation. Reliability Measures 79 where. If the maintenance policy of the Exeter city car rentals is to replace the tyres as soon as it reaches 2000 miles which tyre they should buy: SOLUTION: . After analysing the field failure data of these two tyres it was found that the time to failure distribution of A is exponential with λ = 0. we have ∞ MTTF (t 0 ) = t0 ∫ (t − t 0 ) f (t ) 1 dt = (t − t 0 ) f (t )dt R( t 0 ) R (t 0 ) t0 ∞ ∫ The above equation can be written as (using integration by parts): ∞ MTTF ( t 0 ) = t0 ∫ R(t )dt R( t 0 ) (3. the above expression can be written as: f (t  t 0 ) = h(t ) × R(t ) R (t 0 ) The expression for MTTF(t0) can be written as: ∞ MTTF (t 0 ) = ( t − t 0 )h( t ) t0 ∫ R (t ) dt R (t 0 ) (3. is the conditional probability that the item survives up to time t.3. R(tt0). Example 3.
1 ) can be found from Gamma function table (see β . is given by: R A (2000) = exp(−0.1.5 σ 200 Thus.2 β 2.3678 Reliability of the car tyre produced by company B for 2000 miles. is given by: R B ( 2000) = Φ( µ − 2000 2000 − 2000 ) = Φ( ) = Φ( 0) = 0. it is advisable to buy the tyres produced by company B.8641 1 1 ) = 2000 ⋅ Γ (1 + ) = 1771.1 dt 0.11 The time to failure of an airborne navigation radar can be represented using Weibull distribution with scale parameter η = 2000 hours and β = 2.0005 × 2000) = 0.20). Find the expected value of the remaining life for this radar. It was told that the age of the existing radar is 800 hours. SOLUTION: Using Equation (3. Example 3. The MTTF(800) can be written as: ∞ ∫ R(t )dt ∞ ∫ R (t )dt − ∫ R(t )dt 0 800 MTTF (800) = 800 = 0 R (800) R(800) 800 MTTF − MTTF(800) = MTTF (800) = MTTF = η × Γ (1 + ∫ exp(−( 2000 ) 0 t 2. RA(2000). RB(2000).3.1 The value of Γ (1 + appendix). Reliability Measures 80 Reliability of the car tyre produced by company A for 2000 miles.
2 MTTF of a maintained system Assume that an item is subject to preventive maintenance after every Tpm units. If the item is restored to ‘asgoodasnew’ state after each maintenance activity. = {1 + R( Tpm ) + [ R (Tpm )] +... 0 Tpm 2 Tpm 2 Tpm 3Tpm where Rpm (t) is the reliability of the item subject to preventive maintenance.... etc. that is.90 ) / 0. expected remaining life of the radar aged 800 hours is 1165. the above integral can be written as: Tpm MTTF pm = ∫ R pm (t )dt + ∫ R pm (t )dt + ∫ R pm (t )dt +.90 2000 Thus MTTF(800) ≈ (1771. ∫ exp( −( 0 t 2 . at Tpm . The expected time to failure. then the reliability function between any two maintenance tasks can be written as: R pm (t ) = R[Tpm ] k R (t ).2 . MTTFpm. 2Tpm ..1 ) dt ≈ 763.763.72 hours Thus..21) Using additive property of integration. 3Tpm .. kTpm ≤ t ≤ ( k + 1)Tpm Using the above expression for Rpm(t) in the integral (3.21) we have: Tpm MTTF pm = ∫ R(t )dt + ∫ R(Tpm ) R(t )dt + ∫ [ R(Tpm )] 0 0 0 Tpm 2 Tpm Tpm 2 R( t ) dt +. 3.3.72 hours. Reliability Measures 800 81 Using numerical approximation.16. (MTTF of an subject to preventive) of the item is given by: ∞ MTTF pm = ∫ R pm (t )dt 0 (3.8641 = 1165.} R (t )dt 0 ∫ .
the MTTFpm converges to that of corrective maintenance.3. the above expression can be written as: Tpm 82 MTTFpm = ∫ R(t )dt 0 Tpm 1 − R( Tpm ) = ∫ R(t )dt 0 F ( Tpm ) (3. Thus. . If MTTFpm ≤ MTTF then. MTTFpm can be used to quantify the effectiveness of the maintenance action. if the time to failure is exponential. Reliability Measures As R(t) ≤ 1. The time to failure of the radar follows exponential distribution with mean life 800 flight hours. SOLUTION: We have: T0 = 500 flight hours and (1/λ) = 800 λ = (1/800) = 0.12 A solid state radar is subject to preventive maintenance after every 400 flight hours. Find the MTTFpm of the radar. preventive maintenance will not improve the reliability of the system. It can be noticed that as the value of Tpm increases.12 shows MTTFpm values of an item for different Tpm whose timetofailure can be represented using Weibull distribution with η = 200 and β = 2. the maintenance will not improve the reliability of the item. Example 3. This example is used to demonstrate this well known fact mathematically. when the item is not as good as new after maintenance).5. Figure 3.00125 400 MTTF pm = ∫ exp(−0.00125 × t )dt 0 1 − exp( −0.00125 × 400) = 800 There is no improvement in the MTTFpm because the time to failure is exponential. If MTTFpm>MTTF.22) Similar logic can be used to derive the expression for MTTFpm when the repair is not perfect (that is. then one can say that the reliability can be improved by carrying out maintenance.
The maintenance manager in charge of operation plan to apply preventive maintenance for both the machines for every 200 hours. The time to failure of machine A can be represented by Weibull distribution with η = 1000 hours and β = 2. so that he can improve the expected time to failure of the machines. SOLUTION: The MTTFpm for machine A is given by: 200 ∫ exp( −(t / 1000) 0 2 ) dt ≈ 5033 hours MTTFpm = MTTF pm exp( − (200 / 1000) 2 ) MTTF for machine A is η × Γ (1 + 1 1 ) = 1000 × Γ (1 + ) = 886. MTTFpm of an item for different Tpm values Example 3. The time to failure of machine B can be represented by Weibull distribution with η = 1000 hours and β = 0. preventive maintenance will improve the mean time to failure of the system.3. Reliability Measures 83 MTTFpm for different Tpm values 2500 2000 MTTFpm 1500 1000 500 0 40 80 120 Tpm 160 200 240 MTTFpm Figure 3.13 A manufacturing company buys two machines A and B.2 hours B 2 Thus for machine A. The MTTFpm for machine B is given by: .12.5. Check whether the manager’s decision is correct.
5 Thus for machine B.5 ) MTTF for machine B is η × Γ (1 + 1 1 ) = 1000 × Γ (1 + ) = 2000 hours B 0.17. In British Standard (BS 3527) MTBF is defined as follows: For a stated period in the life of a functional unit. From definition variance V(t) is given by: V (t ) = E ( t 2 ) − [ E ( t )]2 ∞ = t 2 f ( t ) − MTTF 2 0 ∫ Applying integration by parts: ∞ V ( t ) = 2 tR( t ) dt − MTTF 2 0 ∫ (3. 3. preventive maintenance will decrease the mean time to failure of the system.5 MTTF pm = ∫ exp(−(t / 1000) 0 )dt ≈ 414 hours exp( − (200 / 1000) 0.3 Variance of Mean Time To Failure It is important to know the variance of mean time to failure for better understanding of the item. Thus.16. MEAN OPERATING TIME BETWEEN FAILURES (MTBF) MTBF stands for mean operating time between failures (wrongly mentioned as mean time between failures throughout the literature) and is used as a reliability measure for repairable systems. it is better not to apply preventive maintenance for machine B. Reliability Measures 200 84 0. the mean value of the lengths of time between consecutive failures under stated condition.3. .23) 3.
…. Xn etc. may have quite different ages. In 1987 the US Army conducted a survey of the purchase of their SINCGARS radios that had been subjected to competitive procurement and delivery from 9 different suppliers. X3. Table 3. Each of these components will have a different wear out characteristic governed by a different distribution. To determine these expected values it is necessary to determine the distribution type and parameters. In fact. same choice of components (but different manufacturers) and the requirement set by the Army was MTBF of 1250 hours with a 80% confidence.3. X2 . those. same design. The output of this exercise is shown in Table 3. They wanted to establish how the observed Reliability Inservice compared to that which had been predicted by each supplier (using MILHDBK217). X2 . As soon as an item fails. indeed. MTBF can be predicted by taking the average of expected values of the random variables X1 . Reliability Measures 85 MTBF is extremely difficult to predict for fairly reliable items. However. It is interesting to note that they are all same radio. To find the expected value of the random variable X2 one should take into account the fact that not all components of the item are new and.13). This makes it almost impossible to determine the distribution of the random variable X2 and hence the expected value.6 SINCGARS radios 217 prediction and the observed MTBF Vendor A B C D E F G H I MILHDBK217 (hours) 7247 5765 3500 2500 2500 2000 1600 1400 1000 Observed MTBF (hours) 1160 74 624 2174 51 1056 3612 98 472 Let us assume that the sequence of random variables X1 . appropriate maintenance activities will be carried out. This involves replacing the rejected components with either new ones or ones that have been previously recovered (repaired). which are not new. Majority of the suppliers' observed MTBF was no where near their prediction.6 (Knowles. it can be estimated if the appropriate failure data is available. …Xn represent the operating time of the item before ith failure (Figure 3. it is very rarely predicted with an acceptable accuracy. X3 . 1995). .
etc. or for an extended period. X3. altitudes from zero to 50000 feet (15000 meters) and speeds from zero to Mach 2+. thus one should be careful in using the above relation. The value derived by these type of testing will give the expected value of the random variable X1 . X3. One has to test the equipment with some new and some old components to find the expected values of the random variables X2. to be able to predict failure time distribution in all cases. In practice. Military aircraftengines. usually under ‘ideal’ conditions that attempt to simulate the operational environment. are expected to operate while subjected to forces between 5 and + 9 ‘g’. In practice most of the testing is done on new items with all new components in pristine condition.3. as yet. This is currently done empirically by running a sample of items on test until they fail. the expected value of X1 is quoted as MTBF. The above expression can be used only when sufficient amount of data is available. Reliability Measures 86 Operating X1 Down time Operating X1 Down time Figure 3. then MTBF becomes infinity. To calculate MTBF one should consider the expected values of the random variables X2. for example. In fact. the expected value of X1 will give only the Mean Time To First Failure (as the testing is done on new items and the times reflect the time to first failure) and not the MTBF. If the time to failure distribution of the system is exponential then the MTBF can be estimated using the following equation (recommended by MILHDBK217 and DEFSTAN0040): MTBF = T n (3. T is the total operating period and `n’ is the number of failures during this period. etc. If n = 0.13 operating profile of a generic item The science of failures has not advanced sufficiently. .24) where. Note that the above relation is valid only for large value of T.
05t ) = 0.107 is the B10 life for exponential distribution with parameter 0. 2.10 ⇒ t = 2. Reliability Measures Characteristics of MTBF 87 1. Mathematically percentile life can be obtained by solving the following equation for t: F(t ) = ∫ f ( x)dx = p% 0 t (3.18. MTBF is used to predict steadystate availability measures like inherent and operational availability. For a repairable system. this does not include the time spent at repair facility by the system. Percentile life is now frequently used among aerospace industries as a design requirement. where λ is the scale parameter ( also the hazard function ).107 Thus 2. 2. B10% means the life (time) by which 10% of the items will be expected to have failed.3. The main application of percentile life lies in prediction of initial spares requirement (initial spares provisioning. Applications of MTBF 1. Then from above equation we have: 1 − exp( −0.25) Assume that F(t) is a exponential distribution with parameter λ = 0. IP). MTBF is the average time in service between failures. This is the life by which certain proportion of the population (p %) can be expected to have failed. MTBF = 1 / λ for exponential distribution. 3. and we are interested in finding B10 . The value of MTBF is equal to MTTF if after each repair the system is as good as new. PERCENTILE LIFE ( TTFP OR BP% ) Percentile life or Bp% is a measure of reliability which is popular among industries. . Note that.05.05.
Reliability Measures 88 System Reliability and Maintainability Course material Course Instructor: Professor U Dinesh Kumar Indian Institute of Management Bangalore .3.
Determine the operational profile of each block in the reliability block diagram. 2. First. we study the models that are based on simple probability theory. Throughout the Chapter. This may involve performing failure modes and effect analysis (FMEA). The chapter discusses two approaches that can be used to predict the reliability metrics of the system. It lies within the power of man to make this instrument with all its motion' Leonardo da Vinci In this chapter. 5. Derive the life exchange rate matrix (LERM) for the different components within the system. Systems Reliability 89 Chapter 4 Systems Reliability 'A Bird is an instrument working according to a mathematical law. . module. assuming that the timetofailure distributions of different components within the system are known. for predicting different reliability measures. These models can be used only for nonrepairable items. The second approach is based on Markov models. The models for repairable items will be discussed using the Markov models.4. 3. the word ‘system’ is used to represent the complete equipment and the word ‘item’ is used as a generic term that stands for subsystem. Any reliability prediction methodology using timetofailure approach will involve the following steps: 1. we present methodologies that can be used to evaluate systems reliability using simple mathematical tools. Derive the timetofailure distribution of each block. Construct the reliability block diagram (RBD) of the system. part or unit. 4. Compute reliability function of each block. component.
seriesparallel. of an item is a logical diagrammatic illustration of the system in which each item (hardware/software) within the system is represented by a block. Each block within a RBD can represent a component. Systems Reliability 6. In the following sections we address how to evaluate various reliability measures for different reliability block diagrams. The structure of a RBD is determined by the effect of failure of each block on the functionality of the system as a whole. failure of any one item of the system will cause failure of the system as whole. 1 2 … n n Figure 4. water network and Internet). RBD forms a basis for calculation of system reliability measures. 4. A block does not have to represent physically connected hardware in the actual system to be connected in the block diagram.g. The RBD can also have network structures (e. all the consisting items of the system should be available or functional to maintain the required function of the system. The RBD of a hypothetical system whose items are connected in series is given in Figure 4. Each block within a RBD should be described using timetofailure distribution for the purpose of calculating system reliability measures.1. RELIABILITY BLOCK DIAGRAM Reliability block diagram. Construction of RBD requires functional analysis of various parts within the system. Compute the reliability function of the system.1. RELIABILITY MEASURES FOR SERIES CONFIGURATION In a series configuration. subsystem. communication systems. 90 4.20. Thus.4. module or system. parallel. Items whose failure alone cannot cause system failure are connected in parallel. routofn or complex configuration. a RBD can be represented by a series. Reliability block diagram of a system with series configuration Reliability function of series configuration . Depending on the item.19. Series configuration is probably the most commonly encountered RBD in engineering practice. In an RBD the items whose failure can cause system failure irrespective of the remaining items of the system are connected in series. RBD.
which will be discussed later in this chapter. miles.. we use the following equation to find the reliability of the series system. R s = P[ TTF1 ≥ t1 .2). then the reliability function of system for ‘t’ hours of operation is given by: RS (t) = P [ TTF1 ≥ t. Assuming that the random variables TTFi are independent of each other. the reliability of a series configuration with n items is given by: Rs (t ) = ∏ Ri (t ) i =1 n (4. It can have different units such as hours. Systems Reliability 91 Reliability function of a system with series configuration can be derived from the reliability function of its consisting items. In most case time actually represents age or utilisation of the item under consideration. When the life units of items are different (or different items have different utilisation). . However.. TTFn ≥ t ] (4.. it is assumed that the connecting media (such as solder joints) between different items is 100% reliable (unless this is specifically included in the RBD). landings. TTF2 ≥ t 2 . TTF2 ≥ t. One method of normalising the different life units of the items is using Life Exchange Rate Matrix (LERM).4. One has to normalise the ‘time’ before calculating the reliability function in such cases.L .2) Note that in the above equation (4.1) clearly states that the system under consideration will maintain the required function if and only if all the n items of the system are able to maintain the required function for at least t hours of operation. × P[ TTFn ≥ t ] = R1(t) × R2(t) × . cycles etc for different items. In the equation (4.. If TTFi is the timetofailure random variable for the item i.. the expression (4.1) can be written as: Rs (t) = P[TTF1 ≥ t ] × P[TTF2 ≥ t ] × . TTFn ≥ t n ] = R1 ( t1 ) × R2 ( t 2 ) ×L× Rn ( t n ) That is.2) time t is used as a generic term. .. × Rn(t) Thus.. this need not be true.. Let RS(t) represent the reliability function of a series system with n items.1) The equation (4. Let Ri(t) denote the reliability function of the item i.
The value of the reliability function of the system. the item i should survive up to ti. Characteristics of reliability function of a series configuration 1.2 µ = 800 hours σ = 350 η = 2000 hours β = 1.3) unless otherwise specified.3)..n 2.1. for the system to survive up to age t. Table 4. then the system reliability of a series system can be written as: R s (t ) = ∏ exp( − ∫ hi ( x) dx i =1 0 t n n t = exp( − ∫ [ ∑ hi ( x)]dx 0 i =1 Example 4.3) RS (t ) = ∏ Ri (t i ) i =1 n In equation (4.001 η = 1200 hours β = 3. Find the reliability of the system for 500 and 750 hours of operation. Throughout this book we use equation (4.2. which is equivalent to age t of the system. The time to failure distribution and their corresponding parameter values are given in Table 4. That is. Systems Reliability 92 (4. RS(t). That is: RS (t ) ≤ Min {Ri (t )} i =1.75 . ti is the age of the item i.4. for a series configuration is less than or equal to the minimum value of the individual reliability function of the constituting items. each of them are necessary to maintain the required function of the system.1 Time to failure distribution and their parameter of the items Item Time to failure distribution Parameter values Item 1 Item 2 Item 3 Item 4 Exponential Weibull Normal Weibull λ = 0. If hi(t) represent the hazard function of item i.1 A system consists of four items..
the reliability function of the system is given by: Rs (t ) = exp( −0.1 0 1000 1200 1400 1600 1800 2000 2200 2400 2600 2800 3000 3200 3400 3600 3800 4000 200 400 600 800 0 Rs(t) R3(t) R2(t) R4 (t) R1(t) Time Figure 4.7 0.3 0.1759 .75 ) ] R4 ( t ) = exp[ − ( 2000 R2 ( t ) = exp[ − ( Since the items are connected in series.2 800 − t t 1.5 0.2 Reliability function of the system and its constituent items.4 0. the reliability function of various items can be written as: R1 ( t ) = exp( −0.9154 = 0.001 × t ) t 3.1. we get: R(500) = 0.75 ) ] × Φ( ) × exp[ − ( ) ] 1200 350 2000 1 0.2 0.001 × t ) × exp[ − ( t 3.4.4202 R(750) = 0.5568 × 0.4723 × 0. Substituting t = 500 and 750 in the above equation. Systems Reliability SOLUTION: 93 From the information given in Table 4.6065 × 0.8355 = 0.9410 × 8043 × 0.8 0.2 ) ] 1200 800 − t R3 ( t ) = Φ( ) 350 t 1.6 0.9 Reliability function 0.8003 × 0.
4. Systems Reliability
94
Figure 4.2 shows the reliability function of the system and various items of the system. Note that the system reliability value is always less than or equal to any of the constituting items. Example 4.2 Avionics system of an aircraft consists of digital autopilot, integrated global positioning system, weather and ground mapping radar, digital map display and warning system. Apart from the above items, the avionics system has control software. The timetofailure distributions of various items are given in Table 4.2. Find the reliability of the avionics system for 100 hours of operation if all the items are necessary to maintain the required function of the avionics system. Table 4.2 Timetofailure distribution of various items of the avionics system Item Digital autopilot Integrated global positioning system Weather and ground mapping radar Digital map display Warning System Software SOLUTION: From the data given in Table 4.2, we can derive the reliability function of various items as follows: 1. Reliability of digital autopilot Timetofailure distribution Exponential Weibull Weibull Normal Normal Exponential Parameter values λ = 0.003 η = 1200, β = 3.2 η = 1000, β = 2.1 µ = 800, σ = 120 µ = 1500, σ = 200 λ = 0.001
R1 ( t ) = exp( −λ × t ) ⇒ R1 (100) = exp( −0.003 × 100) = 0.7408
4. Systems Reliability 2. Reliability of integrated global positioning system.
95
R2 (100) = exp( −(t / η) β ) ⇒ R2 (100) = exp( −(100 / 1200) 3.2 ) = 0.9996
3. Reliability of weather and ground mapping system radar
R3 (100) = exp( − (t / η) β ) ⇒ R3 (100) = exp( − (100 / 1000) 2.1 ) = 0.9920
4. Reliability of digital map display R4 (100) = Φ( µ −t 800 − 100 ) ⇒ R4 (100) = Φ( ) = Φ(5.8) = 1 σ 120
5. Reliability of warning system R5 (100) = Φ( µ −t 1500 − 100 ) ⇒ R4 (100) = Φ( ) = Φ( 7 ) = 1 σ 200
6. Reliability of software
R6 ( t ) = exp( −λt ) ⇒ exp( −0.001 × 100) = 0.9048
Thus, the reliability of the avionics system for 100 hours of operation is given by:
R s (100) =
∏ Ri (100) = 0.7408 × 0.9996 × 0.9920 × 1 × 1 × 0.9048 = 0.6646
i =1
6
Hazard function of a series configuration Let RS(t) denote the reliability function of the system. From definition, the hazard rate of the system, hS(t), can be written as:
hS ( t ) = −
dRS (t ) 1 × dt RS (t )
(4.4)
Using equation (4.2), the expression for RS(t) can be written as:
4. Systems Reliability
RS (t ) =
96
∏
i =1
n
Ri ( t ) =
∏ [1 − Fi (t )]
i =1
n
(4.5)
where Fi(t) is the failure function of the item i. Differentiating the above expression for reliability function with respect to t, we get:
dR (t ) =− f i (t ) [1 − Fi ( t )] dt j =1 i =1
j ≠i
∑
n
∏
n
(4.6)
Substituting equation (4.6) in equation (4.4), we get
hS (t ) =
∑ Rii (t ) =∑ hi (t )
f (t )
i =1 i =1
n
n
(4.7)
4. Systems Reliability Table 4.3 Hazard rate of series configuration with n items. Probability density function of ith item, fi(t) (Exponential)
97
Hazard function of the system, hS(t)
λi exp( −λi t )
hS (t ) = ∑ λi
i =1
n
(Weibull) β i t β i −1 t ( ) exp( −( ) β i ) η i ηi ηi
n β t h S (t ) = ∑ ( i )( ) βi −1 i =1 η i η i
(Normal) 1 1 t − µi 2 exp( −( ( ) ) 2 σi σ i 2π
n µ −t h S (t ) = ∑ f i (t ) Φ( i ) σi i =1
Thus the hazard function of a series system is given by the sum of the hazard function of individual items. Table 4.3 gives hazard function of a series configuration with n item under the assumption that the timetofailure of the items follows same distribution but have different parameter. Figure 4.3 shows hazard rate of a series system with two items where the timetofailure of individual items follow Weibull distribution.
7 6 5 hazard rate 4 3 2 1 0 0.2 0.4 0.6 0.8 1.2 1.4 1.6 1.8 2.2 Time 2.4 0 1 2
η = 1, β = 2.1
hs(t)
η = 1, β = 0.5
4. Systems Reliability
98
Figure 4.3 Hazard rate of series system with two items with Weibull timetofailure distribution. In most cases, the hazard function of a series configuration will be a increasing function. For example, consider a series system with 10 items. Let 9 out of 10 items be identical and have exponential timetofailure distribution with parameter with rate λ = 0.01. Now we consider two different cases for the timetofailure distribution of the remaining one item.
0.2 0.18 0.16 0.14 Hazard rate 0.12 0.1 0.08 0.06 0.04 0.02 0 105 120 135 150 165 180 195 210 225 240 15 30 45 60 75 90 0
hs(t) h1(t) h2(t)
Time
Figure 4.4 Hazard rate the system with 10 items where 9 of them have constant hazard. Case 1: Let the timetofailure of the remaining one item be represented by using Weibull distribution with scale parameter η = 100 and β = 2.5. Now the hazard rate of this system is given by:
hs ( t ) = 9 × 0.01 + β t β −1 ( ) η η
It is obvious from the above expression that the hazard rate of the system is not constant. Figure 4.4 shows the effect of nonconstant hazard function on the system hazard function even when most of the items have constant hazard function. In Figure 4.4, h1(t) represents the hazard rate for the nine items with exponential timetofailure and h2(t) represent the hazard rate of the item with Weibull timetofailure distribution.
This result proved by Drenick (1961) may not be true for today’s highly reliable systems. This problem will be further discussed in Chapter 8.5. one has to be very careful in using constant hazard function and thus exponential time to failure for complex systems.12 0.08 0.06 0. Thus. Note: The hazard function of complex repairable system may converge to a constant hazard function under certain conditions (mainly under steadystate conditions).01 + β t β −1 ( ) η η It is obvious from the above expression that the hazard rate of the system is not constant.5.04 0.5 shows the effect of nonconstant hazard function on the system hazard function even when most of the items have constant hazard function. h1(t) represent the hazard rate for the nine item with exponential timetofailure and h2(t) represent the hazard rate of the items with Weibull timetofailure distribution. In Figure 4.02 0 15 30 45 60 75 90 0 105 120 135 150 165 180 195 210 225 240 hs(t) h1(t) h2 (t) T im e .4. Systems Reliability 99 Let the timetofailure of the remaining one item can be represented by using Weibull distribution with scale parameter η = 100 and β = 0. Now the hazard rate of this system is given by: hs (t ) = 9 × 0. 0. Figure 4.1 hazard function 0.
8) we have. Example 4.002 + ( η η 760 760 Substituting t = 100 and t = 500 in the above equation. The timetofailure of item A follows exponential distribution with parameter λ = 0. Substituting Ri (t ) = exp( −λ i t ) in equation (4.7 )( ) hS (t ) = h A (t ) + h B (t ) = λ + ( )( ) β −1 = 0.4.5 Hazard function of the system with 10 items where 9 of them have constant hazard.0036 Mean time to failure of a series configuration The mean time to failure. normal. Find the hazard rate of this system at time t = 100 and t = 500.00254 hS(500) = 0. MTTF. The timetofailure of item B follows Weibull distribution with parameter η = 760 and β = 1.002. can be written as: MTTFS = ∫ RS dt = ∫ ∏ Ri (t )dt 0 0 i =1 ∞ ∞ n (4.7. However. 1 − exp( − λ i t ). in case of exponential distribution the expression for system MTTFS can be obtained as follows. hS(100) = 0.7 t 0.8) The above integral can be evaluated using numerical integration if the failure distribution is Weibull. . lognormal or Gamma. hS(t) is given by: β t 1. SOLUTION: Let hA(t) and hB(t) represent the hazard rate of item A and B respectively. denoted by MTTFS. Systems Reliability 100 Figure 4. Assume that the timetofailure distribution of component i is given by. the hazard rate of the system. Since the items are connected in series. of a series configuration.3 A system has two items A and B connected in series.
2 η2 = 20. β2 = 3 . M −1 h × ( R[0] + R[ M * h]) + ∑ h ×R[i × h] 2 i =1 MTTFS ≈ (4.8) can be used whenever the timetofailure of at least one item is nonexponential. Compare the value of MTTFS with mean time to failure of individual items. The timetofailure distribution and their corresponding parameter values are given in Table 4. the value of M is selected such that RS ( M × h) is almost zero. β1 = 2. Note that this result is true only when the timetofailure distribution is exponential.10) Where h is a small value (e.9) Thus.g. 0. Example 4.01 or 0.4 Timetofailure distribution of different items Item Item 1 Item 2 Item 3 Distribution Weibull Exponential Weibull Parameter values η1 = 10.4. The following equation derived using trapezium approximation of equation (4. Find the mean time to failure of the system. Table 4. the MTTFS of a series configuration with n items where the timetofailure of the items are represented by exponential distribution is given by the inverse of the system’s hazard function. Systems Reliability 101 ∞ n ∞ 0 n MTTFs = ∫ ∏ Ri (t )dt = ∫ ∏ exp( −λi t ) dt = ∫ exp( − ∑ λi t ) dt 0 i =1 0 i =1 i =1 ∞ n MTTFs = 1 i =1 ∑ λi n (4.1).5 λ = 0.4 A system consists of three items connected in series.4.
.4. n Min { MTTFi } Where MTTFi denote the mean time to failure of the item i. MTTFS ≤ i =1.5 t ) ) × exp( −0. where MTTFi is the mean time to failure of the item i. .86 System MTTFS ≈ 3. Note that the mean time to failure of the system is always less than that of the components when the items are connected in series. Thus. the MTTFS is given by: MTTFS ≈ 3.48 Table 4.5 gives the mean time to failure of various items. The MTTFS ≤ MTTFi .48 Characteristics of MTTFS of series system 1.2. Table 4.. Systems Reliability SOLUTION: Mean time to failure of the system is given by: MTTFS = ∫ ∏ Ri (t )dt t t = ∫ exp( −( ) β1 × exp( −λt ) × exp( − ( ) β 2 )dt η1 η2 0 0 i =1 ∞ ∞ 3 102 MTTFS = ∫ exp( −( 0 ∞ t 2 .2t ) × exp( −( ) 3 ) dt 10 20 Using numerical integration. the mean time to failure of a system with series RBD will be less than the mean time to time failure of any of its constituting items..87 Item 2 MTTF = 5 Item 3 MTTF = 17..5 Comparison of MTTF of individual items and MTTFS Item 1 MTTF = 8.
flying hours.j]. Thus. rn .. rn. Life exchange rate matrix (LERM) is a square matrix of size n. . Thus the aircraft used in Japan lands more often than the one in USA. This means that the usage of landing gears. For complex repairable systems. . some items of the system may have to operate more than t hours (in many cases it can be less than t hours).. miles. LERM = .. . .1 r 2. Let us denote the life exchange rate matrix as R = [ri. for a system with n items connected in series. It is very common that different items within a system may have different life units such as hour. In this section we introduce the concept of life exchange rate matrix. r2.2 . rn. For example. if the actual mission period is t hours. . the engine may have to operate for more than 10 hours.n . the average flight of a domestic flight within Japan is around 30 minutes compared to that of around 3 hours in US. An aircraft jet engine will be switched on at least 20 minutes before the actual flight.j denotes that: . Thus.4. for 10 hours flight. .21. Systems Reliability 103 2.2 . where n is the number of items in the system. tyres etc of aircraft used in domestic flights in Japan will be much higher than that of USA. .n ..j is the (i.. Thus.2 r2.n The elements of LERM are interpreted as follows: ri.j) th element in the LERM. landings. . In some cases.1 . .1 r1. represents the mean time to first failure. LIFE EXCHANGE RATE MATRIX Not all the components of the item will have the same utilisation or life unit. cycles etc. where ri. 4. Operational environment can also change the ageing pattern of different components within a system. the LERM can be represented as: r1. to find the reliability of a system whose items have different life units it is necessary to normalise the life units.. which can be used to describe the exchange rates between various life units. r1. MTTFS.
ri . let us consider a system with three items connected in series (Figure 4. Using the above matrix. j.j × 1 life unit of j. one can easily measure . The LERM for the above matrix is: 1 10 5 R = 1 / 10 1 0. 2 and 3 be hours.i As an example. Series system with three items where each item has different life units Assume that: 1 hour = 10 miles 1 hour = 5 cycles Using the above data. j = 1 r j . Systems Reliability 1 life unit of i = ri. Any LERM will satisfy the following conditions: 104 ri .5 1/ 5 2 1 One can easily verify that the above matrix satisfies all three conditions for a life exchange rate matrix.6. 1 Hours 2 Miles 3 Cycles Figure 4.k × rk . miles and cycles respectively.6). Let the life unit of items 1.4. k ri . j for all i.i = 1 for all i. it is easy to construct the life exchange rate matrix for the above system. j = ri .
8126 4. PARALLEL CONFIGURATION In a parallel configuration the system fails only when all the items of the system fail. reliability of the system for 5 cycles is given by R1(1)×R2(1)×R3(5).8174 = 0.9942 ) ) = exp( −( 100 η R B (t B ) = Φ ( µ − tB 400 − 240 ) =1 ) = Φ( σ 32 RC (t C ) = exp( −λ × tC ) = exp( −0.5 Reliability block diagram of a system consists of three modules A. to maintain the required function only one item .6.22.00015 per mile. Systems Reliability 105 reliability characteristics in normalised life unit. The timetofailure of module B follows Normal distribution with parameter µ = 400 cycles and σ = 32 cycles. The timetofailure of module A follows Weibull distribution with scale parameter η = 100 hours and β = 3. The timetofailure of module C follows exponential distribution with parameter λ = 0.4. In other words. the module B performs 12 cycles and module C performs 72 miles. B and C connected in series.2 ) ) = 0. Example 4. SOLUTION: For the system to survive 240 cycles. The reliability of individual modules are given by: R A (t A ) = exp( −( tA β 20 3. module A should survive up to 20 hours and module C should survive up to 1440 miles. during 1 hour.8174 The system reliability for 240 cycles is given by: RS ( 240) = R A ( 20) × R B ( 240) × RC (1440) = 0.2. It was also noted that. For the RBD shown in Figure 4. Find the probability that the system will survive up to 240 cycles of module B.9942 × 1 × 0.00015 × 1440) = 0.
the pilot would normally be expected to divert to the nearest airport). Assuming independence among different items.11) where TTFi represents the timetofailure random variable of item i.12) . However..7 1 2 n Figure 4. TTF2 ≤ t . Hence.7 Reliability block diagram for a parallel configuration Parallel components are introduced when the reliability requirements for the system are very high.. the failure function.TTFn ≤ t ] (4. the number of parallel items required should be carefully determined and if possible optimised. Reliability function of parallel configuration Reliability function of a parallel configuration can be obtained using the following arguments. parallel items will increase cost. The reliability block diagram for a system consisting of items connected in parallel is shown in Figure 4. The use of more than one engine in aircraft is one of the obvious examples of parallel configuration (In practice an aircraft would not be allowed to fly if any of the engine fails.. of the system is given by: FS (t ) = P[TTF1 ≤ t . complexity and weight of the system. FS(t). Systems Reliability 106 of the system is required to function. As the system fails only when all the items fail. the above expression can be written as: FS (t) = F1(t) × F2(t) × … × Fn(t) (4. If an engine fails during a flight.4.
RS(t) ≥ Max {Ri (t )} i=1..R2(t) ] × … × [1 . is more than reliability of the any of the consisting items.[1 . RS(t).FS (t) = 1 .. Substituting Fi(t)=1 Ri(t) in equation (4..13) Now.R1(t) ] × [ 1 . Find the reliability of flight control system for 1000 hours of operation.n 2.R2(t)] × … × [1 Rn(t)] or R S (t ) = 1 − ∏ [1 − Ri (t )] i =1 n (4.Rn(t) ] (4. The timetofailure of FCSE can be represented by Weibull distribution with scale parameter η=2800 and β = 2.. RS(t). the reliability function. of a parallel configuration can be written as: RS(t) = 1 .12). the expression for failure function of a parallel configuration can be written as: FS (t) = [ 1 . The system reliability. then the reliability function of a parallel configuration can be written as: R S (t ) = 1 − ∏ [1 − exp( − ∫ hi (t ) dt ] i =1 0 n t Example 4. If hi(t) represent the hazard rate of item i. SOLUTION: Reliability function for a parallel system with four identical items is given by: .14) Characteristics of a parallel configuration 1. That is. Systems Reliability 107 where Fi(t) is the time to failure distribution of item i.6 A flybywire aircraft has four flight control system electronics (FCSE) connected in parallel.4.8.R1(t)] × [1 .
17) Where. we get h S (t ) = {− n d [1 − ∏ (1 − Ri (t ))]} × dt i =1 1 [1 − ∏ (1 − Ri (t ))] i =1 n (4. Example 4.999991 Hazard function of a parallel configuration Hazard function. hS(t).15) Substituting the expression for RS(t) from equation (4. of the parallel configuration can be written as: h S (t ) = − dRS (t ) 1 × dt R S (t ) (4. For t = 1000.4. fi(t) is the probability density function of item i.9455 Thus the reliability of flight control system for 1000 hours of operation is given by: R S (1000) = 1 − [1 − 0.8 ) = 0.7 .9455] 4 = 0.i ≠ j n n 1 − ∏ [1 − Ri (t )] i =1 n (4. R(t) is given by: R (t ) = exp( −(t / η ) β ) = exp( −(1000 / 2800) 2.16) It is easy to verify that the above equation can be written as: ∑ { f j (t ) × ∏ Fi (t )} h S (t ) = j =1 i =1. Systems Reliability 108 RS (t ) = 1 − ∏ [1 − Ri (t )] i =1 4 = 1 − [1 − R(t )]4 where R(t) is the reliability function of each item.14) in the above equation.
can be written as: MTTFS = ∫ RS dt = ∫ {1 − ∏ [1 − Ri (t )]}dt 0 0 i =1 ∞ ∞ n (4. Then the mean time to failure of the system. However. MTTFS.18) For most of the failure distributions one may have to use numerical integration to evaluate the above integral. find the hazard function of the system at time t = 100.5. we get hS(t) = 8. denoted by MTTFS.15)): hS ( t ) = 4 × f (t ) × [ F (t )]3 1 − [ F (t )] 4 where. in case of exponential distribution we can get simple expression for system’s MTTF.8 Mean time to failure of parallel configuration The mean time to failure of a parallel configuration. the hazard rate of the system can be written as (using equation (3. Assume that the timetofailure distribution of component i is exponential with mean (1 / λi ) . is given by: . Systems Reliability 109 For the flight control system electronics discussed in the example 3.0 × 10 .4. SOLUTION: Since all the four items are identical. β t β −1 t ( ) exp( −( ) β ) η η η t F (t ) = exp( −( ) β ) η f (t ) = Substituting t = 100.
Systems Reliability 110 ∞ 0 n MTTFs = ∫ ∏ Ri (t )dt = ∫ {1 − ∏ [1 − exp( −λi t )]}dt 0 i =1 i =1 ∞ n (4. 2. at least r items out of the total n items should maintain their required function for the system to be operational. the expression for MTTFS can be written as: MTTFS = ∫ {1 − [∏ [(1 − exp( −λi t ))]}dt 0 i =1 ∞ 3 1 1 1 1 1 1 1 = + + − − − + λ1 λ2 λ3 λ1 + λ 2 λ1 + λ3 λ2 + λ3 λ1 + λ 2 + λ3 (4. Most of the telecommunication system can be represented as a routofn systems. For the successful completion of the mission.19) For particular values of n.19) can be written as: ∞ MTTFS = ∫ {1 − [(1 − exp( −λ1t )) ⋅ (1 − exp( −λ2 t ))]}dt = ∫ [exp( −λ1t ) + exp( −λ2 t ) − exp( −(λ1 + λ2 )t )]dt 0 0 ∞ = 1 1 1 + − λ1 λ2 λ1 + λ2 Case 2: Assume n = 3. at least three of them should maintain the required function and also the output from at least three programs should agree with each other. ROUTOFN SYSTEMS In an routofn (or routofn:G) system. Equation (4. The reliability function of routofn system can be derived as stated below. Following are few examples of routofn systems: 1.23. we can simplify the above integral to derive the expression for the MTTFS. Case 1: Assume n = 2. Control software in a space shuttle has four programs. .4. This is an example of a 3outof4 system.20) 4.
if the items are not identical then one may have to use other mathematical models such as enumeration to evaluate the reliability. 1.4. r . r. Then the reliability function for the system can be written as: RS(t) = P [{E1 ∩ E2} ∪ {E1 ∩ E3 } ∪ {E2 ∩ E3 }] By putting. For example consider a 2outof3 system with nonidentical items. r .2. = Rn(t). The reliability function of the system can be derived as follows. Systems Reliability 111 Reliability function of an routofn system Consider an routofn system with identical items. is given by: n n R S (t . Weibull timetofailure distribution n n t t RS (t .n). n) = ∑ [ R (t )]i [1 − R (t )] n −i i =r i (4. B = E1 ∩ E3 and C = E2 ∩ E3. n) = ∑ [exp( −λt )]i [1 − exp( −λt ] n−i i =r i 2. A = E1 ∩ E2 .P(A ∩ B) . n) = ∑ [exp( −( ) β )]i [1 − exp( −( ) β )]n − i i η η i = r However. That is.P(B ∩ C) + P(A ∩ B ∩ C) . the above expression can be written as: RS(t.. Let Ei denote the event that the item i successfully completes the mission (or survives t hours of operation).21) For the cases when the timetofailure distribution is exponential or Weibull we have the following expressions for reliability function.r. Exponential timetofailure distribution n n R S (t .P(A ∩ C) .. Then the system reliability.3) = P [ {A ∪ B ∪ C } ] = P(A) + P(B) + P(C) . RS(t. R1(t)=R2(t)=.
n) = + R S (t .n). However. r . n) = ∫ R S (t . The reliability function of r1outofn and routofn system with identical items satisfies the following relation: n r −1 n −r +1 R s (t . MTTF. of an routofn system. then the above integral reduces to a simple expression.4. 3) = R1(t) R2(t) + R1(t) R3(t) + R2(t) R3(t) . there are several approaches available to tackle complex routofn systems with nonidentical items. if the timetofailure distribution is exponential.2 P( E1∩ E2∩E3 ) 112 Let Ri (t) represent the reliability function for the item i.22) Mean Time to Failure of routofn Systems The mean time to failure. Systems Reliability = P(E1∩E2 ) + P(E1∩ E3 ) + P(E2 ∩ E3 ) . r − 1. Now the above expression can be written as: RS (t. The reliability function of 2outof3 system with exponential items are given by: 3 3 R S (t ) = ∑ [exp( − λt )]i [1 − exp( −λt )] n −i i =2 i = 3 exp( −2λt )(1 − exp( − λt )) + exp( −3λt ) Now the MTTFS is given by. . However. 2. For example. n) r − 1[ R (t )] [1 − R(t )] (4. consider a 2outof3 system with identical items where the timetofailure distribution of the item is represented by exponential distribution with parameter λ. n) dt 0 ∞ One may have to use numerical integration in most of the cases to evaluate the above integral. can be obtained using the following expression: MTTFS ( r .2 × R1(t) R2(t) R3(t) The above approach becomes complex when the number of items n increases. r . MTTFS(r.
1.2 2.8 Seriesparallel structure with n items subsystem where subsystem i has mi parallel components In Figure 4.j(t) denote the corresponding reliability of the component.n) and MTTFS (r. SeriesParallel Configuration Here the system has a series structure with n items where each item has parallel redundant items. n) r − 1[ R (t )] [1 − R (t )] 0 (4.m 2 n. n) = ∫ dt + MTTFs (r .22).j) represent jth parallel component of the item i.1 n. then the reliability of item i of the system is given by: Ri (t ) = 1 − ∏ [1 − Ri.24.n) (Misra.m 1 2. Assume that item i has mi components in parallel.1 2. Figure 4.2 1.8 shows a seriesparallel configuration.8. Systems Reliability 113 MTTFS = ∫ [3 exp( −2 λt )(1 − exp( − λt )) + exp( −3λt ]dt 0 ∞ = 5 6λ Using equation (4.m n Figure 4.1 1. (i.24) .2 n. 1992): ∞ n r −1 n − r +1 MTTFS ( r − 1.4. we get the following relation between MTTFS(r1. which have wide application in reliability theory. If Ri. SERIES AND PARALLEL CONFIGURATION In this Section we discuss two types of series and parallel structures.23) 4. j (t )] j =1 mi (4. Model 1.
9 shows parallelseries structure.2 2. .mn Figure 4. Since item i has mi components in series.j(t) is the reliability function of the component j in item i.1 2. . An aircraft with more than one engine. j (t ) j =1 mi (4. is a typical example for these type of configuration. Figure 4.2 1. n. the reliability of item i is given by: Ri (t ) = ∏ Ri.25) Model 2. . .26) where Ri.m1 2. n.1 .1 1. n.m2 . Parallel series structure with n subsystems where subsystem i has mi components Assume that the system has n items connected in parallel where each item has components connected in series. .4. j (t )] i =1 i =1 j =1 n n n (4.2 . Now the reliability of the parallelseries system is given by: R S (t ) = 1 − ∏ [1 − Ri (t )] = 1 − ∏ [1 − ∏ Ri . Systems Reliability Now the system reliability can be written as: 114 R S (t ) = ∏ Ri (t ) = ∏ [1 − ∏ (1 − Ri. .9.27) . j (t ))] i =1 i =1 j =1 n n mi (4. . . . ParallelSeries System 1.
for fast military jets. the other is sufficiently. at cruising speed and altitude. Systems Reliability 115 4. it still offers a considerable resistance and.e. in others. the probability of landing safely with no engines is not very high. If an engine fails. of course. produces an inbalance which has to be offset by the rudder and other controllable surfaces all of which means the functional engine has to work considerably harder thus increasing its probability of failure. In . redundancy is a means of maintaining system integrity if critical parts of it fail.4. even then. In some cases this means replicating parts of the system. It should be noted that in normal flight. alternatives are used. And yet. This is based on the probability that even if one of the engines fails that far from land. reliable to make the probability of not reaching a landing site an acceptable risk. A commercial aircraft has to be able to complete a takeoff and landing with one of its engines shutdown but. If the aircraft only had one engine and it failed. the engines are generally doing very little work and usually are throttled back. extended twin engine operations allows certified twinengine aircraft (e. it would normally be windmilled to minimise 'parasitic' drag but. except under very special circumstances.25. at least. Boeing 777 and Airbus 330) to fly up to 180 minutes from a suitable landing site. ETOPS. i.g. no such aircraft would be allowed to leave the departure gate if any of its engines are not functioning. REDUNDANT SYSTEMS In systems.
it had just enough to get to an exmilitary runway (used as a strip for drag racing). but not the nose wheel. to the point where it would seriously reduce the payload and range. Pratt & Whitney 4084s or General Electric GE 90's (instead of the two it currently has) then there would be true redundancy since it needs only two to achieve ETOPS (Extended Twinengine Operations). These aircraft will glide. If the Boeing 777. Had it not been for the 'redundant' wind turbine. Systems Reliability 116 most cases ultimately. Ontario. but even then. none of the instruments will function and. But. This gives the pilots some control. There was a total blackout in the cockpit and.e. however.4. by pure chance. they are fitted with wind turbines that should drop down and start functioning if there is prolonged loss of power. Eventually the wind turbine deployed which gave them enough power for the instruments. with no power. had a total engine failure some 1500 miles from its intended destination. if the engine cannot be relit. probably making the aircraft uneconomical to operate and hence undesirable to the airlines. some 200 miles outside Winnipeg. large airliners are not going to rise on a thermal. the copilot happened to be particularly familiar with this part of Canada. it would add very significantly to both weight and drag. say. the only option is to eject after directing the aircraft away from inhabited areas. out of eleven other pilots. again by pure chance. There are. happened to be an extremely accomplished glider pilot and. it is almost certain even this experienced glider pilot would have crashed killing all on board. the crew nor the passengers have the option of ejecting or baling out if the aircraft suffers a total engine failure (i. to a certain extend but. The pilot. neither the pilot. a number of problems with this design. Firstly. however good the pilot may be. landed safely with no serious casualties. there will be no power assistance for the control surfaces or to deploy the landing gear. The Gimli Glider as it became known. Secondly such an increase in weight and drag would probably mean the normal two engines would . if there is time. For this reason. radio and power assisted controls to work again. With commercial airlines. For several minutes the pilot manhandled the controls and managed to stop the aircraft from loosing height too quickly. all engines fail). All attempts to relight the engines failed simply because it had run out of fuel. A Boeing 767. Unfortunately the aircraft had lost too much height to reach Winnipeg but. on one of its first flights. even when the copilot managed to find a torch (flashlights) all this showed was that none of the instruments were working (being all digital and computer controlled). was fitted with three or four RollsRoyce Trent 800s. There was just enough power to lock the main undercarriage down. who later tried to land the aircraft in the same circumstances on a flight simulator all crashed.
the redundant items may not be functioning simultaneously as in the case of parallel or routofn configurations. to meet the constantly changing demand for electricity from the 'National Grid' it is necessary to keep a number of steam turbines ready to come on stream whenever there is a surge in demand. the items may be functioning simultaneously but one of them may be sharing much higher load compared to the other. in some of the instances. Systems Reliability 117 provide insufficient thrust therefore either more powerful engines would be needed or. It is quite likely that. warm standby and hot standby. let alone an aircraft with its full complement of passengers and crew. The failure of a generator would result in instantaneous reduction in capacity. batteries may switch in instantly to provide emergency lighting and keep emergency equipment. For example. initially one item will be operating and when this item fails. flights would have been diverted from their scheduled destinations to alternatives. It is important that the switch has to maintain its function. On the Boeing 767. e.02 per thousand flying hours (the standard measure in the aerospace industry). Cold Standby System In a cold standby. The standby redundant systems are normally classified as cold standby. In many cases. Petrol and diesel generators would then be started up to relieve the batteries and provide additional power. respirators and monitors running. That is. none of these had led to the loss of a single life. for safety reasons. Failure of the switch can cause the system failure. the redundant part of the system is switched on only when the main part fails.4. And. a builtin switch senses the failure and switches on the first standby item. In the event of a power cut to a hospital. almost certainly. Whenever the main item fails. one item from the redundant items will be switched on to . the extra engines would have to be used rendering them no longer truly redundant. In some cases. The redundant items will be turned on only when the main item fails. In a cold standby system.g. for example. Such types of systems are called standby redundant systems. have been significantly less than the loss of revenue resulting from the reduced payload had truly redundant engines been fitted. a redundant item is switched on only when the operating item fails. which would be rectified by bringing one of these 'redundant' turbines up to full power. The inconvenience to passengers (and airlines) would have cost the airline but. the amount would. the IFSD (In Flight Shut Down) rate after 10 million hours was less than 0.
4.10). The reliability function of this system can be derived as follows (assuming that the switch is perfect): RS(t) = P{The main item survives up to time t} + P{The main item fails at time u ( u < t ) and the standby items survives the remaining interval ( t . the hazard function of the item in standby mode is zero.u ) } Thus. As an example consider a cold standby system with two items where the timetofailure distribution is exponential with parameter λ. Using the equation (4. R S (t ) = R (t ) + ∫ f (u ) R (t − u )du 0 t (4. Systems Reliability 118 maintain the function. In a cold standby. 1 2 Figure 4.29) the expression for reliability function is given by: R S (t ) = exp( − λt ) + ∫ λ exp( −λu ) × exp( −λ (t − u ))du 0 t = exp( −λt ) + λt exp( −λt ) = exp( −λt )[1 + λt ] For a cold standby system with n identical items with exponential timetofailure distribution. the expression for reliability function is given by: .10 Cold standby redundant system Consider a cold standby system with two identical items (see Figure 4.28) where f(t) is the probability density function of timetofailure random variable.
30) is the cumulative distribution of Poisson distribution with mean λt. One can also derive the expression for nonidentical standby units using the arguments presented in equation (4.30).30) Where R1(t) and f1(t) are the reliability function and failure density function of item 1 and R2(t) is the reliability function of item 2. Using equation (4.31).31) can be easily derived from equation (4.4. Systems Reliability 119 n−1 (λt ) i i =0 RS (t ) = exp( −λt ) ∑ i! (4.31a) Warm Standby System . For the nonidentical MTTF is given by: MTTFs = ∑ 1 λi i =1 n (4. the system reliability function is given by: R s = R1 (t ) + ∫ f1 ( x ) R2 (t − x)dx 0 t (4.31) Equation (4. the reliability function of coldstandby system with nonidentical items is with exponential failure time is given by: Rs (t ) = exp(−λ1t ) + ∫ λ1 exp(−λ1 x) × exp(−λ2 (t − x))dx 0 t R s (t ) = exp( −λ1 (t ) + λ1 [exp( −λ 2 t ) − exp( −λ1t )] λ1 − λ 2 The MTTF of a coldstandby system can be evaluated by integrating the reliability function between 0 and ∞. Assume that the timetofailure items 1 and 2 can be modelled using exponential distribution with mean (1/λ1) and (1/λ2) respectively. For a coldstandby system with nonidentical items.29) The equation (4.29). The MTTF of a coldstandby system with n identical units with exponential failure time is given by: MTTF = n λ (4.
Now the reliability function of the system can be written as: RS (t ) = R (t ) + ∫ f ( x) × R s ( x ) × R (t − x)du 0 t (4. in a warm standby.4. Table 4. If ho(t) and hs(t) represent the hazard rate of a operating and standby item respectively. That is.6 gives the various redundancies and the properties of hazard rate.6 Types of standby redundancy and the corresponding properties of hazard rate Type of Redundancy Properties of hazard rate . the hazard function of the standby item will be less than that of the main item. a standby system can deteriorate even when it is not in use. the main item and the standby item will be sharing equal load. The Table 4. Assume that R(t) and R s ( t ) represent the reliability of the item in operating mode and standby mode respectively. Thus. Thus. the redundant item will be sharing partial load along with the main item. and hence will have the same hazard rate. a hot standby can be treated as a parallel system to derive reliability expressions. Consider a system with two warm standby items.32) For a particular case where R (t ) = exp( −λt ) and R s (t ) = exp( −λ s t ) the reliability function of a warm standby system is given by: R S (t ) = exp( −λt ) + ∫ λ exp( −λu ) × exp( −λ s u ) × exp( −λ (t − u ))du 0 t = exp( −λt ) + λ exp( −λt ) (1 − exp( − λ s t )) λs Hot Standby System In a hot standby. Systems Reliability 120 In a warm standby system.
In such cases. Systems Reliability 121 Cold Standby Warm Standby Hot Standby hS(t) = 0 hO(t) > hS(t) hO(t) = hS(t) 4. 2 1 4 3 5 6 Figure 4. 3.7. For example.26.11 are shown in Table 4. COMPLEX RELIABILITY BLOCK DIAGRAMS In many cases.7.12 Item Distribution with parameter values . 2. Reducing a complex reliability structure will involve the following steps: 1. one has to reduce the block to either a series structure or a parallel structure before one can predict the reliability characteristics of the system.11 Reliability block diagram with combination of seriesparallel structures The timetofailure of the six items within the system shown in Figure 4. the reliability block diagram will have complex combinations of series and parallel blocks. Table 4. Repeat step 1 up till the RBD reduces to either a series or parallel structure. Timetofailure of items shown in Figure 4. consider the RBD shown in Figure 4.11. Compute the reliability of resulting RBD. Replace all purely series (parallel) with an equivalent (reliability wise) single block.4.
11b.11a.4 Lognormal µl = 4. The RBD in Figure 4.11 The block A is same as item 1.00125 The reliability block diagram shown in Figure 4. η = 890 hours. β = 2. 4 5 . η = 450 hours.4. where block B is equivalent to the RBD shown in Figure 4.001 Normal µ = 800.11b RBD equivalent to block B in Figure 4.11a Reliability block diagram equivalent to Figure 4.c.11 2 3 4 Figure 4. Systems Reliability 122 1 2 3 4 5 6 Weibull.75 Weibull. σ = 120 Exponential.11 can be evaluated using the three steps explained above. λ = 0.11 can be replaced by a series structure with three blocks as shown in Figure 4. λ = 0. β = 1. The block B is equivalent to RBD shown in Figure 4.12.75 Exponential. A B C Figure 4.5. σl = 0.
The main advantage of cutset approach is that it is easy to program and most of the commercial software for reliability prediction use cutset approach to evaluate the reliability of complex structures.12) is an example of such configuration. C . The wellknown ‘Wheatstone Bridge’ (see Figure 4. A cutset with minimum number of items is called minimal cut set. To find the reliability of such systems one may have to use special tools such as cutset. pathset. the reliability block diagram may have more complex configuration than the series/parallel structure as discussed so far.11 123 The expression for reliability function of the system in Figure 4. Mathematically. then the system will not fail. the set C will be a minimal cut set if for all ci ∈ C. A cutset is defined as the set of items that.11c. .ci is not a cut set.11 is given by: R s (t ) = R A (t ) × R B (t ) × RC (t ) where R A (t ) = R1 (t ) R B (t ) = 1 − [1 − (1 − R2 (t ) ×R 3 (t )) × (1 − R4 (t ))] RC (t ) = 1 − [1 − (1 − R5 (t )) × (1 − R6 (t ))] For some systems. Then. CUT SET APPROACH FOR RELIABILITY EVALUATION Cutset approach is one of the most popular and widely used methods for predicting reliability of complex structure. 4. will cause the system failure. Here C . RBD equivalent to block C in Figure 4. The cut set approach to reliability prediction involves identifying all the minimal cut sets of the system.4.ci represents the set C without the element ci. when failed. In this Section we illustrate the cutset approach for evaluating reliability of complex structures. if the set C is a cut set of the system. enumeration or the conditional probability approach.27. That is if any item of the minimal cut set has not failed. Systems Reliability Figure 4.
C3 = {2. C2 = {1. 2. 1 1 2 4 C1= 2 C 2= 3 C3 = 3 C4 = C 5 5 4 Figure 4. the set C = {1. each cut set can be considered as a parallel configuration. the cut sets C1.12 Bridge network In Figure 4.12 Since the system will fail when at least one minimal cut sets fail. 3. 2 and 3 will cause system failure. the set of items C = {1. since the failure of the items 1.12.3 = {1. Equivalent RBD for minimal cut sets of the system shown in Figure 4.13.13. 5} Since all the elements of the minimal cut set should fail to cause the system failure. 5}. 4} and C4 = {4. the reliability function of the system can be written as: . C3 and C4 represent the following structures shown in Figure 4.12. 3} forms a cut set. 3} is not a minimalcut set since C . 2. Systems Reliability 124 1 4 3 2 5 Figure 4. For the structure shown in Figure 4. 3. 2}.4. the minimal cut sets are given by: C1 = {1. C2. 2} still forms a cut set. However. Thus.
different minimal cut sets can be treated as a series configuration. RC2 (t). 3. If Ri(t) denote the failure function of the items 1.33) 125 where RC1 (t). Since failure of any one minimal cut set can cause system failure. C2. . Systems Reliability RS(t) = RC1(t) × RC2(t) × RC3(t) × RC4(t) (4. 3.12. In general. each cut set can be treated as a parallel configuration. we get the failure function for the complex structure shown in Figure 4. RC3(t) and RC4(t) are the reliability function of the structures represented by the cut sets C1. 2.34).4. 2. C3 and C4 respectively. RC 2 (t ) = 1 − F1 (t ) F3 (t ) F5 (t ) RC 4 (t ) = 1 − F4 (t ) F5 (t ) RC3 (t ) = 1 − F2 (t ) F3 (t ) F4 (t ). 4 and 5. Substituting the above expressions in equation (4. then we have: RC1 (t ) = 1 − F1 (t ) F2 (t ). Identify all the minimal cut sets of the system. Since all the elements of the minimal cut set should fail to cause the system failure. cut set approach involves the following steps: 1.
There are totally eleven items including the external gearbox.8 η = 7 000.8.5 η = 8 000. β = 3 η = 5 000. Systems Reliability 4. Find reliability of the engine for 3000 hours of operation. Table 4.2 η = 20 000. β = 3 η = 12 000. β = 3 η = 25 000. β = 4 η = 7 000. CASE STUDY ON AIRCRAFT ENGINES Aircraft engine is one of the most critical items used in today’s aviation industry. .8. β = 3 η = 5 000. 126 4. β = 3. Timetofailure distribution of various items of the engine Item no.8 η = 11 000. β = 2. 2. In this section. β = 2. β = 3. oil tank and filter.8 We are interested in carrying out the following tasks 1. Draw the reliability block diagram of the engine. we try to address several reliability measures one may like to know about an engine. β = 3 η = 6 500.28.4. β = 2. 01 02 03 04 05 06 07 08 09 10 11 Item LP compressor LP stage 2 stator Intermediate casing HP compressor HP NGV HP turbine LP NGV LP turbine Exhaust mixer External gear box Oil tank and filter Distribution Weibull Weibull Weibull Weibull Weibull Weibull Weibull Weibull Weibull Weibull Weibull Parameter Values η = 15 000. The timetofailure of these items are given in Table 4.
the reliability of various items for 3000 hours of operation is given by: 1. Find the MTTF of different items of the engine and estimate the MTTF of the engine from the MTTF values of the items. Reliability of LP compressor for 3000 hours of operation is given by: . For an engine of age 5000 hours. Since all the item of the engine must maintain their function. find the mission reliability for 1000 hours of operation. Find the MFOPS of the engine for 500 hours of operation for different cycles.14 Reliability block diagram of the engine 2. reliability function for each of these items is given by: the t R (t ) = exp( −( ) β ) η Substituting the values of η and β for various items in the above equation. SOLUTION: 1. 6. Find the MTTF of the engine if all the items are subject to preventive maintenance after every 1000 hours of operation (assume that after maintenance all the items behave as good as new). 4. Find the hazard rate of the engine at t = 3000 and t = 7000 hours. 7. Systems Reliability 127 3. the system will have a series configuration as shown below: LP compressor LP stage 2 stator Intermediate casing HP compressor HP NGV HP turbine LP NGV LP Turbine Exhaust mixer External gearbox Oil tank and filter Figure 4. Since all the items of the system follow Weibull distribution. 5.4.
8 ) ) = 0.9950 20000 9. Reliability of LP stage 2 stator for 3000 hours of operation is given by: R2 (3000) = exp( −( 3000 2. Reliability of LP NGV for 3000 hours of operation is given by: R7 (3000) = exp( −( 3000 2.9799 R3 (3000) = exp( − ( 11000 4. Reliability of HP compressor for 3000 hours of operation is given by: 3000 3. Reliability of exhaust mixer for 3000 hours of operation is given by: .9997 25000 7.8563 7000 8. Systems Reliability R1 (3000) = exp( −( 3000 3 ) ) = 0.2 ) ) = 0.7872 5000 3. Reliability of LP turbine for 3000 hours of operation is given by: R8 (3000) = exp( −( 3000 2.8 ) ) = 0.5 ) ) = 0.9922 R4 (3000) = exp( −( 12000 5.4.9486 8000 6. Reliability of intermediate casing for 3000 hours of operation is given by: 3000 3 ) ) = 0.9920 15000 128 2. Reliability of HP NGV for 3000 hours of operation is given by: R5 (3000) = exp( −( 3000 3 ) ) = 0. Reliability of HP turbine for 3000 hours of operation is given by: R6 (3000) = exp( −( 3000 4 ) ) = 0.
the reliability of the system is given by 0. Systems Reliability R9 (3000) = exp( −( 3000 3 ) ) = 0. Reliability of oil tank and filter for 3000 hours of operation is given by: R11 (3000) = exp( −( 3000 3. Hazard function of the system.15 hazard function for the engine.8662 5000 Using the above values of individual reliabilities. Reliability of external gearbox for 3000 hours of operation is given by: R10 (3000) = exp( −( 3000 3 ) ) = 0.05 Hazard function 0.9243 7000 129 10.02 0.8 ) ) = 0.06 0.04 0.01 0 10000 11000 12000 13000 14000 15000 16000 17000 18000 19000 20000 1000 2000 3000 4000 5000 6000 7000 8000 9000 0 Time R S (3000) = ∏ Ri (3000) = 0.03 0. 3.4451 i =1 11 Figure 4.4. Since all the items of the system follow Weibull timetofailure.9063 6500 11. the hazard function is given by: .
9 gives the MTTF of different items. Table 4.000791 and hS (7000) = 0. Table 4.004796 Figure 4. one can find the MTTF of different items.4. 4. Systems Reliability β t h(t ) = ( )( ) β −1 η η 130 The system hazard function is given by: hS (t ) = ∑ hi (t ) i =1 11 It is easy to verify that the hazard function of the system at t = 3000 and t= 7000 is given by: hS (3000) = 0.9 MTTF of different item of the engine Item LP compressor LP stage 2 stator Intermediate casing HP compressor HP NGV HP turbine MTTF (in hours) 13 395 4 450 9 823 10 800 7 144 22 650 .15 depicts the hazard function for the engine. The expression for MTTF is given by: MTTF = η × Γ(1 + 1 ) β By substituting the values of η and β.
The mission reliability of the engine is given by: MR(t b . The above expression can be evaluated using numerical integration. The approximate values of MTTFpm is: 1000 MTTF pm = ∫ RS (t ) dt 0 1 − RS (1000) ≈ 999. t m ) R (tb ) where tb is the age of the item at the beginning of the mission and tm is the mission duration. Thus. Systems Reliability LP NGV LP turbine Exhaust mixer External gear box Oil tank and filter 6 202 17 800 6 251 5 804 4 525 131 Since the lowest MTTF is 4 450 (LP stage 2 stator). 5. Substituting tb = 5000 and tm = 1000. we have .075 0. t m ) = R (t b .06 ≈ 27. Mean time to failure of a system subject to preventive maintenance is given by: TP MTTF pm = ∫ RS (t ) dt 0 1 − RS (TP ) It is given that the engine is subject to preventive maintenance every 1000 hours of operation.4.0369 6. TP = 1000 hours. the MTTF of engine will be less than 4 450.
1 0.9 0.3 0. Systems Reliability 132 MR(t b . . etc..1 0 10 11 12 13 14 15 16 17 18 Cycle number 19 1 2 3 4 5 6 7 8 9 Figure 4.6 0. The maintenance free operating period survivability. MFOPS.7 MFOPS 0. Figure 4. t m ) = R (5000 + 1000) = R(5000) ∏ Ri (6000) i =1 11 i =1 11 = ∏ Ri (5000) 0. 2.8 0.02369 7..0013 = 0.16 MFOPS value for different cycles for the engine .0548 0.16 shows the MFOPS values for different cycles (note that these values are derived without considering maintenance recovery period MRP).2 0.4. for the engine described is given by: RS (i × t mf ) R S ([i − 1] × t mf ) MFOPS (t mf ) = = ∏ Ri (i × t mf ) i =1 11 ∏ Ri ([i − 1] × t mf ) i =1 11 The above equation can be evaluated for tmf = 500 and for i = 1.5 0.4 0.
There are several ways that designers can provide maximum utility of their product. Maintainability and Maintenance 133 Chapter 5 Maintainability and Maintenance Maintenance is the management of failures and the assurance of availability J Hessburg Maintainability and maintenance has always been important to the industry as it affects the performance as well as the finance. maintenance forms an essential part of airworthiness. Maintenance is the action necessary to sustain and restore the performance. 1999). maintenance costs around 10% of the airlines total cost. civil or military. The annual maintenance cost of production assets in the United Kingdom is estimated in excess of $13 billion. the main objective of the designer is to provide a reliable and safe item at an affordable price. is to provide a fully serviceable aircraft when it is required by the operator at minimum cost (Knotts. One should be lucky to find a smiling customer when the system fails and it takes a long time to recover the functionality. For commercial airlines. maintenance costs money. Operators/users would like their system to be available and safe to operate when required. Obviously. . with $2 billion wasted through inefficient maintenance management practices (Knotts. For aircraft. The main objective of maintenance is to assure the availability of the system for use when required. 1996). as much as fuel and travel agents' commission (M Lam. 1995). Another is to design systems that are quick and easy to repair when they fail. almost certainly. reliability and safety of the item. One way is to build items/systems that are extremely reliable (and consequently will. However. have a higher acquisition cost). The common objective of aircraft maintenance.5.
Some physical design features such as accessibility. Thus. assessment. Maintainability Engineering is rapidly growing in importance because it provides a very powerful tool to engineers for the quantitative description of the inherent ability of their system/product to be restored by performing specified maintenance tasks. as failure and the degradation of performance is inevitable. . complexity and interchangeability affect the speed and ease with which maintenance can be performed. and maintenance resources. and works out methods for their quantification. The maintainability engineering function involves the formulation of an acceptable combination of design features. repair policies. Maintainability and Maintenance 134 Maintenance also accounts for approximately 10% of the organisations’ employees and at least 1015% of its operating costs. for the user it is equally. testability. factors and resources related to the maintenance tasks needed to be performed by the user in order to maintain the functionality of a system. visibility. Maintainability is the scientific discipline that studies complexity. which directly affect maintenance and system support requirements. It also contributes towards the reduction of maintenance costs of a system during its utilisation to achieve optimum life cycle cost. We also recognised that it is almost impossible for any item to maintain its function forever. 29. All the above information is important as it affects the availability and the life cycle cost of the system. or even more important to know: • • • • • • • When and how often maintenance tasks should be performed How they should be performed How many people will be needed What skills they will need and how much training How much the restoration will cost How long the system will be down What facilities and equipment (special and general) will be required. we showed that it is important for the operator/user to know the reliability characteristics of the item. CONCEPT OF MAINTAINABILITY In the previous chapters. One has to apply a scientific discipline to find answers to these questions.5. prediction and improvement.
manpower. personnel skill. For example. at each prescribed level of maintenance and repair. The job . Of course. maintenance elapsed times and maintenance cost. test equipment. the jet pipe had to be disconnected and removed. Gloster Javelin. Maintainability therefore is an inherent design characteristic dealing with the ease. To remove the jet pipe it was necessary for a technician to gain access through a hatch and then be suspended upside down to reach the clamps and pipes which had to be disconnected. safety. using prescribed procedures and resources. the US Department of Defence's MILSTD721C (1966) defines maintainability as: The measure of the ability of an item to be retained in or restored to specified condition when maintenance is performed by personnel having specified skill levels. accessibility is one of the main concerns for many maintenance engineers. accuracy. Before an engine could be changed.1 illustrates an accessibility problem in one of the older twinengine fighter aircraft. and economy in the performance of maintenance functions. several definitions for maintainability can be found. One of the common misperceptions is that maintainability is simply the ability to reach a component to perform the required maintenance task (accessibility).5. Anon (1992) describes maintainability as: The characteristic of material design and installation that determines the requirements for maintenance expenditures including time. Figure 5. Maintainability can be expressed in terms of maintenance frequency factors. Maintainability requirements are defined in conceptual design as part of system operational requirements and the maintenance concept. Maintainability and Maintenance Maintainability studies have the following objectives (R Knotts 1996): 135 To guide and direct design decisions To predict quantitative maintainability characteristics of a system To identify changes to a system's design needed to meet operational requirements In the technical literature. technical data and facilities to accomplish operational objectives in the user's operational environment.
determining whether the system is safe to operate and. as shown in figure 5. Access Hatch to Disconnect Jet Pipe Access Hatch to Disconnect Jet Pipe Aircraft Skin Items for Disconnection Figure 5. To meet these requirements. there are many other aspects to be considered other than accessibility. which modules within the system are at fault. For safety reasons. It is much easier to maintain an item on the bench. with tools in his hand. The technician had to work his way down between the engine and the aircraft’s skin. testability (ability to detect system faults and fault isolation). It is now widely accepted that false removals (often referred to as No Fault Found – NFF) cost about the same as an actual failure when the component under investigation is removed and replaced. than at the airport gate. the items were outside of the technician’s field of view. amongst busy morning traffic. Maintainability and Maintenance 136 could only be achieved by touch.e.1 Accessibility concern in the Javelin fighter aircraft Another area to be considered under maintainability is troubleshooting the various modules within the allowed time.1 (source: R Knotts). Maintainability should also consider factors such as visibility. that is the ability to see a component that requires maintenance action. Additionally decisionmakers have to be aware of the environment in which maintainers operate. For the commercial airlines. However. if not. i. an easily manageable device is needed which can diagnose with a high degree of accuracy. whereas for a racing car or weapon system every second could be vital. . there is usually less than an hour at the gate prior to the aircraft’s departure to the next destination.5. what action is needed. simplicity and interchangeability. in a war. he was held by his ankles. or in any other resultoriented and scheduledriven environment.
is largely determined at an early stage of the design phase. Devices with these capabilities have been developed in the aerospace.1 Maintenance ElapsedTime The length of the elapsed time. which allows maintainers. we discuss the maintainability measures and how these measures can be used for effective maintenance management. and the cost of maintenance. safety. manhours required completing a maintenance task. However. physical location of the item. safety of the restoration. novice as well as expert. etc). For example. 30. would be a big cost saver. accessibility of the items. It will. also be influenced by other factors during the various stages of the life of the system but any bad decision made (either explicitly or by default) during . Maintenance elapsed times are even advertised as a marketing strategy. Measures of maintainability are related to the ease and economy of maintenance such as. For similar purposes the Flight Control Division of the Wright Laboratory in the USA has developed a fault detection/isolation system for F16 aircraft. cost. not just the mean time but also the probability that a maintenance task will be completed within a given time. or even more important to have information about the characteristics with which to define the maintenance time. it is equally. tools. It is therefore a function of the maintainability and supportability of the system. MEASURES OF MAINTAINABILITY It is extremely important for the user to have information about the functionality. therefore. required for the restoration of functionality. as well as the decisions related to the requirements for the maintenance support resources (facilities. Formula 1 racing car and luxury car industries. As the elapsed time has a significant influence on the availability of the system. and other characteristics of the product under consideration at the beginning of its operating life. trained personnel. testability.5. to find failed components. the Boeing 777 includes an 'onboard maintenance system' with the objective to assist the airlines to avoid expensive gate delays and flight cancellations. The maintenance elapsed time is influenced by the complexity of the maintenance task. elapsed time that an item spends in the state of failure. In the next section. called time to restore. operators would like to know the maintenance times. 30. frequency of maintenance. Maintainability and Maintenance 137 Reducing the number of false removals. of course. spares.
Mean Time to Repair One approach for measuring maintainability is through Mean Time to Repair (MTTR). one can use the cumulative distribution function of the elapsed time to find the percentage of maintenance tasks that will be completed within a specified time. Conditional factors which represent the influence of the operating environment and the consequences of failure with the physical condition. Consider an arbitrarily large operating time T. responsibility and other similar characteristics related to the personnel involved. experience. lighting. that is. the expected number of failures of unit i in during T is given by: T MTTFi The mean of total repair time to repair unit i during T is given by: (5. motivation. noise. training. Environmental factors which represent the influence of factors such as temperature. physical ability. selfdiscipline. Personnel factors which represent the influence of the skill. 2. 1. 1994). Maintainability and Maintenance 138 the design stage will be costly to rectify at a later stage and will significantly affect both the operational costs and system availability.1) . With the knowledge of the reliability and maintainability of the subsystems one can evaluate the maintainability of the system. vibration. and others such as those similar to the maintenance personnel factors during restoration. time of the day. mean time to repair of the system. This maintainability measure can be represented using the probability that the maintenance task considered will be completed by a stated time. Assuming that the failure rate of the unit is constant. wind. MTTRs (Birolini.5. humidity. and shape of the item under restoration. MTTR is the expected value of the item's repair time. time of the year. attitude. Assume that the reliability block diagram of the system has a series structure with n items with no redundancy. Let MTTFi and MTTRi be the mean time to failure and mean time to repair of subsystem i in the system. geometry. Since the maintenance elapsed time is a random variable. noise. 3.
the mean number of failures is given by: T i =1 MTTFi ∑ For the whole system.6) The MTTF and MTTR of four subsystems in a system are given in Table 5. that is.1 (5. ∑ n MTTR i (5.3) and (5.3) i =1 ∑ MTTRi × n T MTTFi (5. as: MTTFi MTTRs = i =1 n 1 ∑ i =1 MTTFi Assuming constant failure rate. MTTRs. Maintainability and Maintenance 139 (5. the mean of total repair time is given by: n (5. Subsystem 1 2 MTTF 200 500 MTTR 24 36 . equation (5. we get the mean time to repair at the system level. MTTRs .1.4) Combining equation (5. Estimate the system level mean time to repair.5.4).5) can be written as: MTTFi i =1 n λ i MTTR i λs i =1 MTTRs = ∑ Example 5.2) MTTRi T MTTFi For the whole system.5) λi = n 1 and λ s = ∑ λi .
5. Maintainability and Maintenance 3 4 SOLUTION: Applying equation (5.5), we get: 340 420 12 8
140
24 36 12 + + + MTTRs = 200 500 340 1 1 1 + + + 200 500 340
8 420 ≈ 20 hours. 1 420
Mean Time to Repair MultiIndenture Case Many complex systems are broken down into a number of levels of indenture (LoI). For these systems, recovery of an LoIi unit is usually achieved by the removal and replacement of LoIi+1 items. In many cases, the replacement LoIi+1 item will not be the item that was removed. It may be a new (i.e. unused) one or it may be one that was removed from another LoIi unit and subsequently recovered and put into stock for such an occasion. Now, for such a system, the time to repair will be the time to remove and refit the units at the next lower level of indenture. The elapsed time will need to take into account logistic delays (i.e. waiting for equipment, personnel, spares and any transport to and from the site at which the maintenance work is to be done). This is discussed in more detail in Chapter 10. Suppose a system is made up of n levels of indenture and a unit at LoIi is made up of mi LoIi+1 items. Suppose also that to recovery an LoIi unit, one of the mi items is removed and replaced with average times, MTTRMi,j and MTTRPi,j respectively. Let us assume that the probability that item j is rejected given that unit i has been removed is Pi , j then over an arbitrarily long operating time T, the expected number of system failures is:
T MTTF1
where MTTF1 is the mean time between failures of the system (over time T). Now, the probability that the failure was due to subsystem j is Pi,j so the mean time between failures due to subsystem j is
5. Maintainability and Maintenance
141
MTTF1, j =
1 1 MTTF1 = = P1, j λ 1, j P1, j λ 1
Assuming the system reliability block diagram is series and is series and there are no redundancies. The expected number of failures of subsystem j is
P, j 1
T T = = λ 1, j T MTTF1 MTTF1, j
The expected time to recover the system given that subsystem j is the cause of its failure is
MTTR1, j = MTTRM1, j + MTTRP , j 1
The expected total time spent recovery the system due to subsystem j failures over time T is then
P, j 1
MTTR1, j MTTF1
T=
MTTR1, j MTTF1, j
T = λ 1, j MTTR1, j T
So, the expected total time spent recovering the system by subsystem exchange is
∑P
j =1
m1
MTTR1, j MTTF1
1, j
T=∑
j =1
m1
MTTR1, j MTTF1, j
T = ∑ λ 1, j MTTR1, j T
j =1
m1
Where m1 is the number of subsystems. Then the mean time to recover the system (by subsystem exchange per system failure) is
MTTR1, E = ∑
j =1
m1
λ 1, j λ1
MTTR1, j
To determine the total maintenance time, we would have to look at the time spent recovering the subsystems, by subsubsystem exchange and so on down to the lowest level components that are recovered in this way and
5. Maintainability and Maintenance
142
then add on any time spent repairing the lowest level components (parts) if they can be repaired but we will leave this exercise until our next book.
30.2
Maintenance Man Hour (MMH)
Although elapsed time is an extremely important maintenance measure, one must also consider the maintenance manhours, MMH (also known as maintenance labour hours). The MMH is an estimate of the expected “spannerinhand” time and takes into account all of the maintenance tasks and actions required for each system, subsystem or component recovery. It should be noted that the MMH can be considerably greater than the elapsed time as it is often possible and sometimes even necessary to employ more than one person on any given activity or task. “Work study” and “time and motion” exercises have generated tables of times for every conceivable maintenance action, from releasing the catches that are used on access panels to inspecting the blades on a turbine using a boroscope to drilling out a stud that has sheared after too much torque has been applied to it, to disconnecting and reconnecting all of the pipes and leads when removing and replacing an engine. In most cases, these times are based on carrying out these tasks and actions in ideal conditions, i.e. in a properly lit workshop, which is heated and provides shelter from the elements. They are generally done when the components are in pristine condition free from contamination, corrosion or damage. It is also generally assumed that the mechanic carrying out each action will have been properly trained and familiar with the correct procedures. In practice, however, it is very rare for all of these ideal conditions to be met so, the actual times will inevitably be longer than those used in the MMH prediction. Maintenance manhours are useful in their own right but very often they are given as a “rate” such as (MMH/operating hour), (MMH/cycle), (MMH/month), and (MMH/maintenance task). For example, elapsed times can be reduced (sometimes) by increasing the number of people involved in accomplishing the specific task. However, this may turn out to be an expensive tradeoff, particularly when high skill levels are required to perform the tasks. Also, unless it actually requires more than one person to do the job, there is likely to be an “interference factor” which means that the efficiency of each person is reduced. Therefore, a proper balance among elapsed time, labour time, and personnel skills at a minimum maintenance cost is required. Commercial airlines and air forces use the measure Maintenance ManHour per Flight Hour (MMH / FH) as an indicator of the maintainability of the
5. Maintainability and Maintenance
143
aircraft for comparison with other similar aircraft either of an older generation or made by another manufacturer. This measure may be used to decide between alternatives although, in many cases, it will be used to exert pressure on the manufacturer to make improvements. The following expression can be used to evaluate the MMH/FH:
MMH / FH =
N1 (t ) × MPMT × MNC pm + N 2 (t ) × MCMT × MNC cm Total flying hours
(5.7)
Where: N1(t) is the total number of preventive maintenance tasks during t hours, and N2(t) is the total number of corrective maintenance tasks. The value t should be equal to the operational life of the aircraft. MPMT = Mean preventive maintenance time. MCMT = Mean corrective maintenance time. MNCpm = Mean number of crew for preventive maintenance. MNCcm = Mean number of crew for corrective maintenance. Note that these estimated mean values should be weighted according to the expected frequency of each maintenance task as we did when calculating MTTRs above. A problem with estimating the MMH/FH metric is that it relies on the reliability of the various components of the system, which may be agerelated and will, inevitably, depend on the maintenance and support policies. For these reasons, the MMH/FH may not remain constant with aircraft age. The implication of using such a metric is that it is preferential for it to be minimised, however, it may actually be both cheaper and yield a higher level of availability if more time is spent on maintenance, particularly preventative maintenance.
30.3
Maintenance Frequency Factors
Maintainability engineering is primarily concerned with designing a system so that it spends a minimum time in maintenance, given that it needs maintaining. Another characteristic of system design pertaining to maintainability is in optimising the mix between preventative and corrective maintenance. The ideal system design would allow the operators to use the system until just before it fails but, with enough notice of the impending failure so that
5. Maintainability and Maintenance
144
the operator can choose to perform the necessary maintenance at the most opportune moment. In all but a few cases, prognostics have, as yet, not reached this level of sophistication. An alternative approach is builtin redundancy and faulttolerant systems. These allow the operators to defer maintenance for a limited period or, in certain circumstances until the backup system fails. Corrective maintenance can be expensive if the failure causes damage to other parts of the system or if it stops the system from earning its keep. However, redundant components will also add to the cost of the system and may reduce its loadcarrying capacity. The spare wheel in cars takes up space that could otherwise be used for carrying luggage, it also increases the gross weight, which will reduce the performance of the car both by reducing its rate of acceleration and increasing the fuel consumption. It is common practice for motorists to replace tyres before the tread has been completely worn away because it is unsafe to drive on bald tyres. It is also illegal and the penalties can be both expensive and inconvenient. It is also very easy to inspect tyres for wear so it is possible to leave them until the “last minute” or get them replaced when the car is not needed thus minimising the inconvenience or lack of availability. Brake pads are more difficult to inspect by the owner. As a result, many cars are now fitted with pads that have an inbuilt electrode, which causes a warning light to be illuminated on the dashboard when it comes into contact with the metallic disc (due to the nonconductive part of the pad being worn away). this generally gives the driver a sufficient warning for him or her to find out what the warning light means and take the necessary corrective action before the brakes become dangerous. Most motorists have their cam or timing belts replaced within about 1000 miles of the manufacturer’s recommended mileage possibly during a routine service (scheduled maintenance) or at the driver/owner’s convenience. In this case, the owner has almost certainly no way of knowing how much longer the belt will last and, indeed, it is likely to cost them almost as much to have the belt inspected as it would to have it replaced because of the amount of work involved. In this case, the extent of the damage to the engine if the belt breaks is likely to cost a great deal more than that of replacing the belt early. It would no doubt be possible to devise a monitor that could indicate when the belt was starting to wear but, whether it would be practical in terms of its size, reliability, cost and extra weight is very much open to debate. Here we have seen four different solutions to the same problem of avoiding failures and hence the need for corrective maintenance. One of the tasks
Maintainability and Maintenance 145 of the maintainability engineer is to determine which. training. If a component is repaired then it is likely that the time to failure for that component will be less than if it had been replaced by a new one. which is the sum of all costs related to elements of logistics support which are required to perform the considered maintenance task. one way of measuring maintenance cost is cost per maintenance task. or other similar approaches is appropriate taking into consideration the costs and practicalities in each circumstance.5. 30. the frequency with which each maintenance action must be performed is a major factor in both corrective and preventive maintenance. Obviously this is greatly influenced by the reliability of the components but it can also be related to the type and frequency of the maintenance performed. Preventative maintenance may cause components to be replaced unnecessarily (or at least prematurely). a clean environment equipped with expensive. Personnel and human factor considerations are also of prime importance. skill level and number of technicians. We will return to the question of repair effectiveness in Chapter 6. If a maintenance task requires highly skilled personnel. Support considerations cover the logistics system and maintenance organisation required to support the system. Prognostics can help but these too have their own problems of reliability and the need for maintenance as well as possibly adding to the weight. that it will prove economical to perform this task at first line or. even at second line.4 Maintenance cost factors For many systems/products. maintenance costs constitute a major segment of the total lifecycle cost. experience has indicated that maintenance costs are significantly affected by design decisions made throughout the early stages of system development. In addition to the above factors. . They include the availability of spare parts. technical data (manuals). Thus. Maintainability is directly concerned with the characteristics of system design that will ultimately result in the accomplishment of maintenance at minimum cost. Further. special tools then it is unlikely. test equipment and required special and general tools. possibly. complexity and cost of the system. if any of these. These considerations include the experience of the technician. There is clearly a need to strike a balance. Allowing a system to run until it fails may maximise the times between maintenance but failures can be expensive to rectify both because of the extent of the damage caused and because of the loss of availability of the system whilst it is being maintained.
5. If such replacements can be performed within the 50 min. This allows the mechanics to prepare to replace these items as soon as the aircraft has reached the gate. compilation of maintenance manuals. etc. it was not so much a question of removing the engine from the aircraft as removing he aircraft from the engine. Some of the early jet fighters were virtually built around the engine so that. For example. Generally. The demonstration is also expected to generate results that can contribute to the whole development process. Maintainability and Maintenance 146 However. Any maintainability demonstration would involve the following steps: . If. which signal ahead to the destination any detected faults in the mission critical components (i. A recent innovation on commercial aircraft is to use autonomics. identifying any remaining deficiencies such as the design of the system and the test equipment.e. Anyone who has seen the film Battle of Britain or Reach for the Sky will recognise the importance of turning fighter aircraft around in minimum time when the airfield may be under attack from enemy bombers and fighters. will mean that the engine will either have to be replaced or overhauled/reconditioned before it is likely to function again. if the maintainability engineer had designed the system so that this task could be done by personnel with lower skill levels using standard tooling then it might have allowed the task to be done in the field with a possible reduction in the turnaround (or outofservice) time. as a result of a failed cam belt. those not on the minimum equipment list). in order to replace the engine. 31. turnaround time then it will not be necessary to find a replacement aircraft or delay the departure. An aircraft not in the air is bit like a duck out of water. the most important issue is whether the system can be recovered by subsystem (or line replaceable unit – LRU) exchange within the specified times. MAINTAINABILITY DEMONSTRATION The objective of the maintainability demonstration is to show that the various maintenance tasks can be accomplished in the times allotted to them. It is a common requirement that each LRU can be removed and replaced without interfering with any other LRU. there is little to be gained by making it easy to replace a broken cam belt by the side of the road. The damage done to the engine. it is particularly vulnerable and do very little to defend itself. the task is only likely to be done once in the system’s life during a major overhaul when it would be at a central maintenance unit or returned to the manufacturer then such considerations may be less relevant. or so.
Maintainability and Maintenance 147 1. Simulate the system failures and perform corrective maintenance action. we have to show that: MTTR * ≤ MTTR + zα s n (5.α) 100 percent confidence limit is given by: MTTR + zα s n (5. training.8) Where zα is the z value (standard normal statistic) that locates an area of α to its right and can be found from the normal table.645. Once we have the recorded repair time data from the above procedure. Let t1. 2. then it is easy to verify whether the maintainability target has been achieved using the following procedure. the (1 . maintenance manuals. A mix of repairers representative in skills. Identify the operation and environmental condition in which the system is likely to be used. 3. the equipment/tools. then to demonstrate that the system has achieved this. …. For n > 30. MTTR and 's' are given by: MTTR = 1 n 1 n 2 2 ∑ t i .5. t2. tn denote the observed repair times to complete the repair tasks for a sample of n units. and experience of those who would do the actual repair in service must conduct the repair. for a 95% confidence limit. it is an important to take care of the following issues during the demonstration: 1. Further. One should also record the maintenance manhours required to complete the repair task successfully. The test conditions must be representative. 2. and s = ∑ (t i − MTTR ) n i =1 n − 1 i =1 If the target maintainability is MTTR*.9) . The test must be on a sample of fixed final build standard. lighting and similar factors must be carefully considered. For example. the zα is given by 1.
05. the condition for acceptance is given by: MTTR * ≤ MTTR + tα . If the target MTTR is 20 hours. s.n1 can be obtained from the tdistribution table shown given in appendix.2. Example 5.2.5. s = 19 i =1 20 i =1 From the tdistribution table (see appendix) we get. The MTTR and standard deviation.2 A maintainability demonstration test is carried out on 20 parts and the accomplished repair times are shown in Table 5. n −1 s n (5. we use tdistribution. n1 = 1.472 .33 4.45 + 1. Recorded repair times form a sample of 20 parts in hours 8 12 32 10 SOLUTION: Since the observed number of data.06 = 22. Maintainability and Maintenance 148 Whenever the number of repair time data is less than 30. we use tstatistic.10) The value of tα.45 hours. check whether the system has achieved the target maintainability using 95% confidence level. n1 = 19). Table 5. are given by: 6 9 26 14 12 17 30 32 20 4 19 26 24 40 10 18 MTTR = 1 n 1 20 2 ∑ (t i − MTTR ) = 10.06 hours ∑ t i = 18. in that case. n is less than 30. 95% upper limit for MTTR is given by: MTTR + tα s n = 18. tα.729 (α = 0.729 × 10.
Thus the achieved MTTR is significantly greater than the required MTTR and is therefore not acceptable. all actions.5. 4. by keeping the system in a proper condition for a longer time. In other words. 3. a state in which it can perform a required function. Extend the life of the system. MAINTENANCE CONCEPT The maintenance concept begins with a series of statements defining the input criteria to which the system should be designed. maintenance can be defined as: The combination of all technical and administrative actions. The purpose of maintenance is to keep systems in a state of functioning in accordance with their design and to restore them to a similar state as and when required. 8. 32. 2. which keep the system running and ensure that it is maintained to an acceptable standard in which it is able to operate at the required levels efficiently and effectively. Maintainability and Maintenance 149 Which is greater than 20 hours. In other words. Ensure that the system is fit and safe to use. 33. which is the target MTTR. Reduce the consequences of failure. The objectives of maintenance are to: 1. Maintain the value of the system. 7. Maintain reliability and achieve a high level of safety. or restore it to. Ensure that the condition of the system meets all authorised requirements. to increase the up time of the system. These statements relate to the maintenance tasks that should be performed at each level of . intended to retain an item in. 6. including supervision actions. Maintain the system's availability and therefore minimise production and quality losses. Reduce overall maintenance costs and therefore minimise the life cycle cost. 5. MAINTENANCE According to BS 4778.
the skill levels of the maintenance personnel that perform the identified tasks. 3. lines or [maintenance] levels. LEVELS OF MAINTENANCE Complex systems can be considered as made up of several levels of indenture. and is a prerequisite to system design and development. 2. armament. and anticipated maintenance environmental requirements (Knezevic. the auxiliary power unit (APU) and various accessories including control units and pumps.. It provides a basis for detailing the maintenance plan and impacts upon the elements of logistic support. may be made up of subassemblies and parts. propulsion and general. The propulsion system then becomes a LoI1 item that may consist of the engines. each of which may be considered as LoI2 items. in turn. avionics. al. may be thought of as consisting five subsystems: airframe. personnel quantities and skill levels. spare parts. 1997). At the same time.5. A preliminary maintenance concept is developed during the conceptual design stage. and other resources. facilities. “First . task frequencies and time. 1995] • • • • The anticipated level of maintenance Overall repair policies Elements of maintenance resources The organisational responsibilities for maintenance The maintenance concept serves the following purposes: 1. the military typically divides its maintenance and support infrastructure into 3. Maintenance concept at the design phase tends to ensure that all functions of design and support are integrated with each other. Maintainability and Maintenance 150 maintenance (organisational. An engine is typically an assembly of a number of modules or LoI3 items which. the test equipment and tools that should be used in maintaining the system. is continually updated. LoI4 and 5 respectively. 34. A combat aircraft that may be considered as the Level 0 (LoI0). It provides the basis for the establishment of requirements for total support which include maintenance tasks. It provides the basis for the establishment of maintainability and supportability requirements in the system design. 4 or 5 echelons. intermediate and depot). The maintenance concept evolved from the definition of system operational requirements delineates [Blanchard et. maintenance time constraints.
Some minor repairs and routine servicing may also come under this category. personnel skilllevel requirements. supplier or original equipment manufacturer (OEM) often provides a shadow facility at “Fourth Line” effectively duplicating the Third Line facility’s capabilities. the manufacturer and the user should define a basic repair policy that may vary from repair/replace a part (LoI5. major overhaul.g. The criteria in which the maintenance tasks selected at each level are. generally by module (or shopreplaceable unit – SRU) exchange. repair and replacement of major items in a system. say) to replace the entire system. external adjustments and replacement of line replaceable units (LRU). simple condition and performance monitoring activities. which are performed at workshops (mobile.2 Intermediate level Intermediate maintenance level is related to all maintenance tasks. task complexity. maintaining oil levels. special maintenance equipment and resources and economic measures. Traditionally. These are usually supported by a depot or maintenance unit at “Third Line” or “DLevel”. The hierarchies of achieving maintenance tasks are divided into three or four levels. rearming. a removed LRU would be recovered. etc. Within the scope of the identified level of maintenance. e. system modification. “Second Line” or “ILevel” is typically the main operational bases from which the squadrons are deployed.5. 34. This would include replenishment tasks.1 User level (organisational) This type of maintenance level is related to all maintenance tasks which are performed on the system whilst it is on deployment or at its operating site. Performing maintenance tasks at this level require higher personnel skills than those at organisational level and additional maintenance resources. The contractor. 34. . Common maintenance tasks accomplished at this level are detailed condition and performance monitoring activities. semimobile and/or fixed) where the systems would normally be based. or “OLevel” is from where the systems are operated. Maintenance levels are concerned with grouping the tasks for each location where maintenance activities are performed. at this level. refuelling. Maintainability and Maintenance 151 Line”.
particularly if they require expensive equipment or are likely to take a long time. more appropriate maintenance policies and. etc. which are accomplished beyond the capabilities of intermediate level at remote sites.) should be consistent and more accurate. Such contracts are often funded by fleet hour arrangements such as “powerbythehour”.3 Depot level Depot maintenance level is related to all maintenance tasks. “Third Line” refers specifically to an operatorowned facility whereas in US nomenclature “DLevel” also includes manufacturer/contractor facilities. is that because all of the maintenance is done in one place. The advantage to the operators is that they can get on with what they are in business for. perhaps more strongly by the OEM than the operator.4 HoleintheWall With the move to ever greater efficiency and/or minimal costs. hopefully to the operator. Maintenance tasks at depot level include complete overhauling and rebuilding of the system. In the UK system. reduced logistic delays. Maintenance tasks at this level are carried out by highly skilled specialists at a specialised repair facility or the equipment producer’s facility. the perceived need to reduce manning levels and the desire of OEMs to increase their revenue by entering the “aftermarket”. . and again. see chapter 12. to improved designs.5. They would also include tasks which may only be performed rarely. they (the OEM) are the best people to take it apart and repair it. 34. cause of failure. items repaired or replaced. the people doing it should become more efficient (as they see the same job more often) and the inservice data (time to failure. Maintainability and Maintenance 152 34. It is also argued. etc. putting “bums on seats” or “bombs on target”. highly complex maintenance actions. This is then passed through this mythical hole in the wall to the OEM or maintenance contractor in exchange for a replacement (serviceable) LRU. The contractor then takes the LRU away to a convenient location where it is recovered. that having designed and built the LRU. A secondary advantage to the OEM. the “holeinthewall” concept is gaining in popularity. ultimately. Better data should lead to improved forecasting. This is where the only intrusive maintenance task the operator performs is to remove the LRU (at first line).
In order to achieve this. by performing appropriate maintenance tasks. Figure 5. conditional maintenance task Each maintenance task is briefly discussed in the following sections. It also requires an appropriate environment in which the maintenance activities can be carried out. The execution of a maintenance task requires resources such as the right number and skills of personnel. Maintainability and Maintenance 153 35. Resources Environment Need for maintenance Maintenance task activities Maintenance task complete Restore functionality Figure 5. or termination of the item/system functionality.2 Process of maintenance task Maintenance tasks can be classified into the following three categories: 1.1 Corrective Maintenance Task Corrective maintenance task. at least. as long as they are needed. 35. preventive (predictive) maintenance task 3. which is performed with the intention of restoring the functionality of the item or system. which is initiated by the need for maintenance due to a reduction. in a specified manner. MAINTENANCE TASK CLASSIFICATION All users would like their systems to stay in a state of functioning as long as possible or. after . it is necessary to maintain the system’s functionality during operation. Thus. CRT. material. maintenance task can be defined as a set of activities that need to be performed. is a set of activities.5.2 shows the process of maintenance task. corrective maintenance task 2. equipment. etc. in order to maintain the functionality of the item/system.
4 illustrates the activities of a typical preventive maintenance task.5.3 illustrates typical corrective maintenance task activities. Figure 5. Maintainability and Maintenance 154 the loss of the functionality or performance (i.2 Preventive Maintenance Task Preventive maintenance task. PMT. Corrective maintenance task is also referred to as an unscheduled or unplanned maintenance task. CMT complete CMT start Verification Item Failed Test and Check MTBF Fault Location ?DMT c Disassembly Corrective maintenance task activities Assembly Repair or Replacement Figure 5. represents the elapsed time needed for the successful completion of the task. DMT p . represents the elapsed time needed for the successful completion of the task. The duration of the preventive maintenance task. DMTc. Figure 5.3 Activities of typical corrective maintenance task 35. .e. after failure). is a maintenance activity that is performed in order to reduce the probability of failure of an item/system or to maximise the operational benefit. The duration of corrective maintenance task.
regardless of the actual condition of the items/systems.5. hours). distance (e. .g. renewal and overhaul. These tasks are performed. miles) or number of actions (e. Maintainability and Maintenance 155 PMT complete PMT start Disassembly Verification Tp MTBF DMTp Test and Check Preventive maintenance task activities Replacement Assembly Figure 5. landings).4 Activities of a typical preventive maintenance task Preventive maintenance task is performed before the transition to the state of failure occurs with the main objective of reducing: • • The probability of the occurrence of a failure The consequences of failure Common preventive maintenance tasks are replacements.g.g. at fixed intervals based on operating time (e.
DMT m . through monitoring of some condition parameter(s) it would be possible to identify the most suitable instant of time at which preventive maintenance tasks should take place. COT complete COT start Decision making Inspection/ Examination FMTI/ FMTE Data collection Condition interpretation DMTm Conditional maintenance task activities Condition assessment Figure 5. COT does not normally involve an intrusion into the system and actual preventive action is taken only when it is believed that an incipient failure has been detected. the . COT. represents the elapsed time needed for the successful completion of the task. recognises that a change in condition and/or performance is likely to precede a failure so the maintenance task should be based on the actual condition of the item/system. in recent years.3 Conditional (Predictive) Maintenance Task Conditional maintenance task. In the past.5. corrective maintenance and preventive maintenance tasks have been popular among maintenance managers.5 illustrates the activities of a typical conditional maintenance. Figure 5. The duration of conditional maintenance task. However.5 Activities of a typical conditional maintenance task. Thus. Maintainability and Maintenance 156 35.
5. i. system producers and /or users to achieve high safety. . FBM. may be very convenient but is likely to result in an increase in the amount of maintenance needed because parts will be replaced when they have achieved a fraction of their expected life. Note that the failure could be due to a component of the system that was not being subjected to IBM or it could have happened as a result of some unpredictable external event such as foreign object damage or because the inspection interval was too long or the inspection was ineffective. until the execution of a preventive maintenance task is needed or a failure occurs. LBM. on the other hand. 4) ExaminationBased maintenance policy. where conditional maintenance tasks in the form of inspections are performed at fixed intervals of operation. EBM. Routine or scheduled preventive maintenance. Maintainability and Maintenance 157 disadvantages of these tasks have been recognised by many maintenance management organisations. where preventive maintenance tasks are performed at predetermined times during operation. 3) InspectionBased maintenance policy. IBM. reliability and availability at minimum cost. where corrective maintenance tasks are initiated by the occurrence of failure. The need for the provision of safety. 2) TimeBased maintenance policy. With respect to the relation of the instant of occurrence of failure and the instant of performing the maintenance task the following maintenance policies exist: 1) FailureBased maintenance policy. until the performance of a preventive maintenance task is required or until a failure occurs requiring corrective maintenance. its failure may cause significant damage to other parts of the system and will often occur at inopportune times causing a disruption to the operation and inconvenience to the users. MAINTENANCE POLICIES The maintenance policy defines which type of maintenance will (normally) be performed on the various components of the system. loss of function or performance. at fixed length of operational life. where conditional maintenance tasks in the form of examinations are performed in accordance with the monitored condition of the item/system. It is determined by maintenance engineers. and reduction of the maintenance cost have led to an increasing interest in using conditional maintenance task. Waiting until a component fails may maximise the life obtained from that component but.e.. 36.
5. Corrective maintenance task priorities can range from "normal". fire fighting.6. A schematic presentation of the maintenance procedure for the failurebased maintenance policy is presented in Figure 5. in order to restore the functionality of the item/system considered.1 FailureBased Maintenance Policy FailureBased maintenance policy. "urgent" to "emergency". According to this policy. Maintainability and Maintenance 158 The principal difference between the above maintenance policies occurs at the time when the maintenance task is performed. maintenance tasks often take place in ad hoc manner in response to breakdown of an item following a report from the system user. this approach to maintenance is known as breakdown. The advantages and disadvantages of each maintenance policy are briefly described below. reactive. represents an approach where corrective maintenance tasks are carried out after a failure has occurred. postfailure. These categories reflect the nature of the response rather than the actual actions done. Failure based maintenance could be the most applicable and effective maintenance policy in situations where: Items for which the loss of functionality does not compromise the safety of the user and/or the environment or the failure has little or no economic consequences (i. categories major and minor see FMECA in Chapter 11) systems have builtin redundancy or have been designed to be faulttolerant Operating time Item Failed Figure 5. or unscheduled maintenance. Consequently.e. FBM. 36.6 FailureBased Maintenance Policy Advantages of failure based maintenance .
which is a function of the timetofailure distribution of the item considered and in some cases it may be adjusted by the system's user. The reason for that is the fact that the . It demands a lot of maintenance resources. planned or scheduled maintenance. a timebased maintenance policy is performed at fix intervals.5. by carrying out maintenance actions before failure occurs. Therefore. Disadvantages of failure based maintenance Despite the advantages of implementing this policy.2 TimeBased Maintenance Policy Some failures can lead to economical consequences such as loss of production and therefore a reduction in profit. third parties and environment. passengers. This policy is very often called agebased. As the main aim is to reduce the probability of occurrence of failure and avoid the system breakdown. Maintenance activities cannot be planned. which is the ratio of the Mean Duration of Utilised Life of the item (MDULF ) to the expected operating life (MTTF). it is desirable to prevent these failures. lifebased. if possible. CU. The failure of an item can cause a large amount of consequential damage to other items in the system. of items considered will have value of 1. This means that the noncritical items will have the ability to perform their function(s) for the stated period of time when they operate under stated conditions. The user will get maximum value out the component when the FB maintenance policy is applied. Maintainability and Maintenance 159 Implementation of FBM to the above situations could lead to full utilisation of the operating life of the item. Analysis of maintenance costs have shown that a repair made after failure will normally be three to four times more expensive than the same maintenance activity when it is well planned [Mobley (1990)]. This means that coefficient of utilisation. • • • • The failure of an item will generally occur at an inconvenient time. it has some disadvantages when it is not correctly selected. 36. Some failures may have an impact on the safety of the user.
the condition of the system. loss of life or serious injury) 3.g. for example. The cost of lost production and of consequential damage can be reduced. and potential costly outages avoided. years. A schematic presentation of timebased maintenance procedure is presented in Figure 5. the total costs of applying this policy are substantially lower than the alternatives 4. FMT L . Downtime. which may be based on. preventive maintenance tasks take place. 3. Maintenance can be planned ahead and performed when it is convenient from the operational and logistics point of view. hours. or its consisting items. Maintainability and Maintenance 160 maintenance task is performed at a predetermined frequency. is determined even before the item has started functioning. which could have catastrophic consequences to the user/operator and environment (Chernobyl. the likely consequences of failure is catastrophic (e. Piper Alpha and similar) it may be the only feasible option. miles. the probability of occurrence of failure is reduced 2.7. Safety can be improved. that make it is possible to plan all tasks and fully support them in advance. For failures. Thus. The frequency of maintenance task. the time that the system is out of service. 2.5. which are summarised in the following list: 1. Bhopal. Advantages of timebased maintenance policy One of the main advantages of this maintenance policy is the fact that preventive maintenance tasks are performed at a predetermined instant of time when all maintenance support resources could be planned and provided in advance. operating times such as. can be minimised. . The timebased maintenance policy could be effectively applied to items/systems that meet some of the following requirements: 1. cannot be monitored or is impractical or uneconomical. number of actions or any other units of use. 4. at the predetermined length of operational life specified. Timebased maintenance has many advantages over failurebased maintenance.
7 Time Based Maintenance Policy Disadvantages of time based maintenance policy In spite of the advantages given above. Maintainability and Maintenance FMTL 161 System in use Preventive Task Operating time Predetermined Time Tp Figure 5. 1. If the time to perform the maintenance is greater than the time the system would normally be idle (eg overnight) then because of the frequency. irrespective of their condition. will have a value much less than one. This policy could be uneconomical because the majority of items are prematurely replaced. 4. it could cause higher levels of unavilability. A summary of the disadvantages of timebased maintenance policy is listed below. 6. Reducing the probability of failure by prematurely replacing components means that the coefficient of utilisation of the item/system. 5. It cannot guarantee the elimination of all failures and will do nothing to reduce nonagerelated failures. The tasks may require higher numbers of skilled mechanics. In many industries this policy is now only used under special conditions because it is very costly. Increasing the frequency of maintenance tasks may lead to an increase in the probability of human errors in the form of maintenanceinduced failures. . Consequently. CU L . a large number of unnecessary tasks will be carried out on a system that could have been operated safely for a much longer time. Timebased maintenance is performed irrespective of the condition of the system. 3. and also because its efficiency in reducing failures is not always supported by experience. the timebased maintenance policy has several disadvantages that must be minimised.5. 2.
and has proved its ability to extend the operating life of a system without increasing the risk of failure is conditionbased maintenance.There is a high probability of detecting potentially catastrophic failures (before they happen).8. CBM is also known as predictive maintenance. its performance and/or other condition parameters.8 Condition based maintenance policy . Maintainability and Maintenance 162 36. 3. This policy is worth applying in situations where: 1.The cost of the condition monitoring technique is lower than the expected reduction in overall maintenance costs. A policy which overcomes many of the disadvantages of the previous maintenance policies (failurebased and timebased). 2. This should mean the system is operated in its most efficient state and that maintenance is only performed when it is costeffective. This means that the principle reason for carrying out maintenance activities is the change or deterioration in condition and/or performance. and the time to perform maintenance actions is determined by monitoring the actual state of the system. Conditionbased maintenance can be defined as: "Maintenance carried out in response to a significant deterioration in a unit as indicated by a change in the monitored parameters of the unit's condition or performance" [Kelly & Harris (1978)].The state of the system is described by one or more condition parameters. A schematic presentation of conditionbased maintenance procedure is presented in Figure 5. Inspection time Condition System in use System in use Satisfactory Unsatisfactory Preventive Task Operating time Figure 5. and reduced maintenance costs have led to an increasing interest in development of alternative maintenance policies. CBM. increased system availability.5.3 Condition Based Maintenance (Predictive Maintenance) The need for the provision of safety.
usually through wear but also through damage.g. the SFC may be subject to considerable noise and hence any deterioration will only be apparent by using sophisticated trending algorithms. e. in particular. As the system deteriorates. For systems that are operated in an inconsistent manor for which the environmental conditions may be in a constant state of change. miles per gallon. constant ambient temperature.Performance Trend Monitoring For propulsion or energy producing systems.Examination This is a conditionmonitoring task. consistent changes in the specific fuel consumption (SFC) will almost certainly be indicative of a deterioration in the system which will need some form of maintenance to restore it to an acceptable level. 2. Maintainability and Maintenance 163 The conditionbased maintenance is a condition or performancedriven preventive maintenance. pressure and output). This is possible because of the unique properties and characteristics of the relevant condition predictor. 3 . The principle of conditionbased maintenance therefore is based on the way of monitoring the condition parameters of a system giving three different types of condition monitoring: 1. such as Kalman Filtering. kilometres per litre. .Inspection Inspection is generally performed at regular intervals using any of a number of nondestructive test (NDT) procedures which are designed to determine whether the condition of the (inspected) item is satisfactory or unsatisfactory and hence whether further action is required. The results directly affect the scheduling of the next examination. these ratios may show signs of decreasing. the performance may be expressed as a ratio of the output to input. thrust per kilogram or (mega)watts per tonne. This means that the timing of the maintenance task is not simply a function of the meantimetofailure.g. For systems operating in relatively constant conditions (e.5. which presents a numerical description of the condition of the item at that moment through relevant condition predictors.
Care should be taken to ensure that all of the maintenance significant items are identified and listed.5. especially the consequences of failure. 1995): A. the first step of the conditionbased maintenance decision process is a comprehensive review of all items in a system. Therefore. RCI The Relevant Condition Indictor.3. MSIs. Relevant Condition Indicator. FMECA and Reliability Centred Maintenance. it is necessary to use the following management steps that are shown in Figure 5. Identification and selection of condition parameters Once the maintenance significant items are identified it is necessary to determine all monitorable parameters which describe their condition or performance. The condition parameter can be defined as a measurable variable able to display directly or reflect indirectly information about the condition of an item at any instance of operating time. maintenance engineers would like to find many condition/parameters which can be monitored and which accurately reflects the condition /performance of the system. These are items whose failures could be safetycritical. each item within the system should be analysed from the point of view of failure. in order to identify the maintenance significant items. since it is likely to be both uneconomical and impractical to monitor them all. is a parameter that describes the condition of an item during its operating time and it indicates the condition . Ideally.9 Identification and selection of maintenance significant items The first requirement of implementing CBM is to decide which items of the system should be monitored.1Setting up conditionbased maintenance policy 164 In order to implement CBM policy. Thus. environmentally damaging or revenue sensitive. In practice there are two distinguishable types of condition parameters which are able to achieve this (Knezevic et al. Maintainability and Maintenance 36. RCM (see also Chapters 6 and 11). RCI. The most frequently used engineering tools for performing this task is a Failure Mode Effect and Criticality Analysis.
Maintainability and Maintenance 165 of the item at the instant of inspection. The numerical value of RCI represents the local value of the condition of an item/system at the time of inspection. RCI is not able to predict the future development of the condition of the considered item/system. geometry. which describes the condition of an item at every instant of operating time. This type of condition parameter is usually related to the performance. level of oil. the level of vibration. RCP. It is necessary to stress that the RCI could have an identical value at different instances of operating time. Typical examples of RCP are: thickness of an . etc. Typical examples of the RCI are performance. which describe the condition of the item under consideration. However. and other characteristics. is a parameter. Usually this parameter is directly related to the shape. System/ items selection Condition parameters selection Condition monitoring techniques selection Collecting data & information accepted Condition assessment & fault diagnosis unaccepted Conditionbased maintenance task Figure 5. weight. corrosion fatigue crack growth. pressure. The general principles of the RCP are discussed by Knezevic (1987). temperature. The RCP represents the condition of the item/system which is most likely to be affected by a gradual deterioration failure such as wear.5.9 Flow of condition based maintenance Relevant Condition Predictor (RCP) The Relevant Condition Predictor.
performance. in the end. crack length. These developments have made it possible to obtain more reliable information on the condition of the system. etc are available for use by maintenance engineers in order to determine measurable value of condition parameter. It is important to understand the behaviour of the failure that the item exhibits so that the most effective monitoring techniques can be chosen. Selection of condition monitoring technique Having identified the maintenance significant item and the associated condition parameter(s). With the increased interest in condition monitoring in recent years there have been a number of developments in the techniques that are used to collect data and provide information. the next step is to select the suitable monitoring technique. In many instances such information is used to insure that the status of the system will continue to be in a functioning state without significant risk of . The RCP cannot have identical values at two or more instance of time. the type of condition parameter and. Maintainability and Maintenance 166 item. which helps maintenance engineers assessing the condition of an item or a system. The numerical value of the relevant condition predictor at any instant of operating time quantifies the cumulative value of the condition of an item/system at the time of examination. which will be used to inspect and examine each condition parameter. etc. it is possible to define the equipment or instrument that will be needed to carry out condition monitoring. Once the decision is made as to which techniques are to be used. NDT techniques.5. Collecting data and information The philosophy of condition monitoring is to assess the condition of an item/system by the use of techniques which can range from human sensing to sophisticated instrumentation. Numerous condition monitoring techniques. The condition monitoring technique is a device used to inspect or examine an item in order to provide data and information about its condition at any instance of operating time. for instance. vibration. depth of tyre treads. in order to determine the need for performing preventive maintenance tasks. on cost and safety. The decision as to which conditionmonitoring techniques are selected depends greatly on the type of system.
and in some instances to make a decision on the timing of when maintenance tasks should be performed.5. Condition monitoring and condition assessment Condition assessment . Maintainability and Maintenance 167 breakdown. These may then be analysed by an onboard computer to determine whether there has been a change in the condition of the item/system and whether that change requires any action. The method of data collection can be classified into the following categories: Online data collection and monitoring Online data collection and monitoring uses instrumentation fitted to the system which takes continuous measurements of the condition parameters. or taking a physical sample. Periodic monitoring therefore provides a way of detecting progressive faults in a way that may be cheaper than the online system. Figure 5. The benefit of using online monitoring is to reduce the need for human intervention and minimise the probability of a failure occurring between inspections. This type of method involves either the collection of data using a portable data collector.10. lubrication oil samples for analysis of contamination and debris content. Offline collection and monitoring Offline collection and monitoring is periodic measurement of a condition of an item/system or continuous data collection which is analysed remotely. for example.
Artificial Intelligence techniques such as Expert System. it can detect incipient failures through its online monitoring of the condition parameters of the system [Lavalle et al (1993)]. The volume of data necessary to accurately determine the condition of the item/system can require an excessive amount of time to process and analyse.10) can range from human experience to sophisticated instrumentation. the demand to manipulate and process large amounts of data very quickly has lead to the development of tools such as Artificial Intelligence. it is necessary to have reliable methods of interpreting the data to identify whether the considered item is undergoing a transition from the normal to abnormal condition and in many cases to identify the causes of the changes. to assist engineers to gain maximum value from the data. Consequently. By processing information much faster (than humans) the time to assess the condition and diagnose the causes of failures can be reduced. The last few decades have seen a number of developments in the methods which are used to help the maintenance engineers assess and diagnose the condition of an item/system and provide them with information on which to base their decision. In recent years. These techniques extend the power of the computer beyond the usual mathematical and statistical functions by using dialogue and logic to determine various possible courses of action or outcome. It can analyse situations objectively and will not forget any relevant facts (given that it has been supplied them). therefore the probability of making a wrong assessment or diagnosis may be reduced. Effective conditionbased maintenance requires a large number of measurements taken continuously or at intervals that assure recognition of change in the condition of the item/system in sufficient time to avoid the need for any corrective action. condition based maintenance could be divided in two policies: Inspection Based Maintenance Policy .5. AI. According to the classifications of condition parameter. Maintainability and Maintenance 168 The assessment of the condition of an item/system (Figure 5. Furthermore. Neural Networks and Fuzzy Logic have been applied to the discipline of monitoring and diagnostic systems [Mann et al (1995)]. the aim of this step is to implement condition based maintenance. Once condition monitoring sensors have been installed and data are being collected. Implementation of condition based maintenance Having identified and listed all the condition parameters of the maintenance significant items.
Before the item/system is introduced into service the most suitable frequency of the inspection. is shown in Figure 5.5. By eliminating all unscheduled interruptions to operation and production and only carrying out required maintenance in a carefully controlled manner. If the item fails between inspections. it is possible to reduce the maintenance cost.11 Inspection is carried at fixed intervals to determine whether the condition of the item. The algorithm. improve the efficiency of the operation and increase the system’s availability. FMT I . and critical value of relevant condition indicator RCI cr has to be determined. Advantages of inspection based maintenance CBM has the potential to produce large savings simply by allowing items in the system to be run to the end of their useful life. is satisfactory or unsatisfactory according to the RCI. RCI is inspectionbased maintenance. the prescribed preventive maintenance tasks take place. to improve safety. RCI ( FMT I ) > RCI cr . Maintainability and Maintenance 169 The suitable maintenance policy for items for which their conditions are described by the relevant condition indicator. corrective maintenance takes place. . which presents the maintenance procedure in this case. This reduces the equipment down time and minimises both scheduled and unscheduled breakdown situations. Once the critical level is reached.
Reduce maintenance resources due to reduction in unnecessary maintenance activities 6.5. Improve availability by being able to keep the system running longer and reducing the repair time. since monitoring and detection of the deterioration in condition and/or performance of an item/system will enable the user to stop the system (just) before a failure occurs.11 Algorithm for inspection based maintenance task The benefits of inspection based maintenance policy can be summarised as follows: 1. since maintenance engineers can determine optimal maintenance intervals through the condition of constituent items in the system. This allows for better maintenance planning and more efficient use of resources. Improve safety. Extending the operating life of each individual items and therefore the coefficient of life utilisation will be increased compared to time based maintenance 4. Reduce unplanned downtime. The above benefits will lead to a reduction in maintenance costs . 5. 3. Maintainability and Maintenance 170 MAINTENANCE PROCEDURE Conditionbased Maintenance policy Type Inspection FMT and RCI Determination of I cr System in use Inspection of RCI cr RCI > RCI Yes Maintenance task No Figure 5. 2.
With more information about the process of change in condition. Therefore. It is a dynamic process because the time of the next examination is fully determined by the real condition of the system at the time of examination. RCPinitial < RCP(l ) < RCPcr : continue with examinations. Examination based maintenance provides additional information about the change in condition of the items considered during its operational life. as in the case of timebased preventive maintenance but with fuller utilisation of operating life.: 1. 3. according to the numerical value of the RCP . The interval between the limit ( RCPlim ) and critical values depends on the ability of the operator to measure the condition of the item through the RCP . Dynamic control of maintenance tasks allows each individual item to perform the requested function with the required probability of failure. RCPcr < RCP(l ) < RCPlim : preventive maintenance task required. However. examination based maintenance approach is proposed by Knezevic (1987b) for the determination of maintenance tasks based on relevant condition predictors. . The item under consideration could be in one of the following three states. RCPlim < RCP(l ) : corrective maintenance task. Consequently. Maintainability and Maintenance 171 Examination Based Maintenance Policy The decision for performing the conditionbased maintenance tasks is based on the information related to the condition of an item/system established through condition checks during its operational life. due to an increased number of interruptions of the operation caused by increasing the number of inspections. 2. sets the limit above which appropriate maintenance tasks should be performed.5. The critical level of the relevant condition predictor RCPcr . examination based maintenance was developed for the control of maintenance procedures [ElHaram 1995]. because the failure has already occurred. as an alternative. This indicates that inspectionbased maintenance strategy has achieved the demand for increasing the level of utilisation of an item/system. hence with a reduction of total cost of operation and production. the system availability may not increase. a higher level of utilisation of the items can be achieved whilst maintaining a low probability of failure during the operation.
FMTnE . dependent on the value recorded: which means that a prescribed maintenance task should take place. 2. which immediately arises here. either of the two conditions is possible. MRCP ( FMT1E ) < RCPcr . FMT1E . and it presents the real condition of the item at this instant of time. The following two conditions are possible. MRCP ( FMT1E ) > RCPcr .5. The greater the difference. 4. the longer the (operational) time to the next examination. the item can continue to be used. . Provision of the required reliability level of each individual system as in case of timebased maintenance. 1. 2. Fuller utilisation of the functional life of each individual system than in case of time based maintenance. Maintainability and Maintenance 172 In order to minimise interruptions to the operation and maximise the availability of the system. At the predetermined time of the next examination. is: when will the next examination have to be done. and the same procedure should be followed. FMT2E . as shown in Figure 5. preserving the required reliability level? The time to the next examination depends on the difference between the RCPcr and MRCP ( FMT1E ) . Reduction of the total maintenance cost as a result of extending the realisable operating life of the system and provision of a plan for maintenance tasks from the point of view of logistic support. The question. Increased availability of the item by a reduction of the number of inspections in comparison with inspectionbased maintenance.12 Advantages of Examination Based Policy The advantages of the examinationbased maintenance policy are: 1. 3. no stoppages occur until the time to the first examination of the condition of the item. MRCP ( FMT1E ) . The result of the examination is given as a numerical value of the relevant condition predictor.
Maintenance Policy Condition Based Maintenance Policy Type Examination Determination of FMT1E and RCP cr System in use Examination of RCP at FMT1E MRCP ( FMT1E ) > RCPcr Preventive task Determination of FMTnE System in use Examination of RCP at FMTnE MRCP ( FMT nE ) > RCPcr Figure 5. . conditionbased maintenance should not be considered to be a standalone policy. Thus.5. the optimal selection of maintenance policy for a system should include failurebased. Maintainability and Maintenance 173 5. It should be integrated as a part of the overall maintenance policy. Applicability to all engineering systems. it may not be economical or practical to use examinationbased maintenance. In some cases. The main difficulties are the selection of a relevant condition predictor and the determination of the mathematical description of the RCP(l ) .12 Maintenance procedure for examination based maintenance In practice. Sometimes it is not physically possible to monitor the condition of all maintenance significant items. For these reasons. it is impossible to eliminate all breakdowns.
The reasons for this are summarised below: 1. metrology and calibration equipment. 3. equipment. MR. consumables. testability. A maintenance management approach such as reliability centred maintenance could be used to select the most applicable and effective maintenance task for each item in the system 37. MP: required for the installation. Resources required primarily to facilitate the maintenance process will be called Maintenance Resources. Typically. 2. MTE can be divided into two groups: special to type equipment (STTE) and general (to type) equipment (GTTE). inspection based and examinationbased maintenance strategies. The time required to perform a maintenance task will also depend on decisions made during this phase.. MTE: includes all tools. Maintainability and Maintenance 174 timebased. the most suitable policy is examinationbased maintenance. MSS: is generic name which includes all spares. accessibility and any special facilities. tools or resources needed. It may not be possible or practical to monitor the condition / performance of the significant items. their sequence and the type and quantity of resource required mainly depends on the decisions taken during the design phase of the item/system. Maintenance Test and Support Equipment. maintenance stands and servicing and handling equipment required to support maintenance tasks associated with the item/system. the suitable maintenance policy is therefore. The resources needed for the successful completion of every maintenance task. special supplies.5. Not all items in the system are significant. Maintenance Personnel. handling. repair items. failurebased maintenance. timebased maintenance. and related inventories needed to support the maintenance process 2. so the suitable maintenance policy is therefore. diagnostic and checkout equipment. could be grouped into the following categories (Knezevic 1997): 1. checkout. For significant items with relevant condition predictors. such as the complexity. 3. If the condition parameters of a significant item cannot be described by a relevant condition predictor. then the suitable maintenance policy is inspectionbased maintenance 4. and sustaining maintenance of the item/system and its . MAINTENANCE RESOURCES It is important to stress that the number of activities. special condition monitoring equipment. Maintenance Supply Support.
Maintenance Facilities. Platform on which maintenance task is performed (on operational site. MCR: refers to all computer equipment and accessories. Such data not only cover the system but test and support equipment. necessary in the performance of maintenance functions. drawings and specifications that are necessary in the performance of system maintenance functions. facilities information. humidity. modification instructions. working environment . Maintenance Technical Data. This includes both condition monitoring and diagnostics. solar radiation. and similar). data bases and so on. accuracy and ease of task completion. we call it maintenanceinduced failure MIF. maintenance instructions. program tapes/disks. portable buildings. transportation and handling equipment. real estate. and special repair and overhaul facilities must be considered related to each maintenance task 5. training. The main environmental factors could be grouped as follows: • • • space impediment (which reflects the obstructions imposed on maintenance personnel during the task execution which requires them to operate in awkward positions) Climatic conditions such as rain/snow. improper use of test equipment. workshops. inspection and calibration procedures. on board a ship/submarine. accuracy and ease of task completion. training equipment and facilities 6. dry dock. MTD: necessary for checkout procedures. software. 38. Maintainability and Maintenance 175 associated test and support equipment are included in this category. space vehicle. Formal training for maintenance personnel required for each maintenance task should be considered 4. inspection pits. MAINTENANCE INDUCED FAILURES Whenever the cause of failure is related to the maintenance performed on the system. housing. On the other hand. Maintenance Computer Resources. it is important to remember that each task is performed in a specific work environment that could make a significant impact on the safety. which could make significant impact on the safety. overhaul procedures. temperature. calibration laboratories. and similar situations.5. The root cause of MIF is poor workmanship. maintenance shops. MFC: refers to all special facilities needed for completion of maintenance tasks. Physical plant. which might lead to poor spares or material selection.
When the cabin was pressurised as the aircraft climbed to cruising altitude.5. if they are not “captured” then there is a danger of them being “lost” when undone. the window blew out. In 1983. of course. The result was that the wheel overtook the car as Nigel was exiting from the pit lane and his chance of victory and of the championship ended at that moment. In 1991. Nigel Mansell lost his chance of becoming the Formula 1 World Champion in Portugal when one of the mechanics during a routine tyre change crossthreaded the retaining nut on the rear offside wheel. Fasteners not properly tightened and locked (where appropriate) can work loose. a new Air Canada Boeing 767 flying from Montreal to Edmonton ran out of fuel half way between the two at Gimli near Winnipeg. The window was removed and replaced during a recently completed maintenance activity. They are also ones where it was relatively easy to determine the cause(s). their miscalculations in converting between imperial and metric units was the final straw in an unfortunate sequence of events. the design team are aware of these needs and their importance to the operational effectiveness of the aircraft. Using CATIA and EPIC (or similar systems) can do a great deal to aid the task of making components accessible and removing interference provided. A combination of his size and the quick reactions of other members of the crew were all that saved him from a certain death. Maintainability and Maintenance 176 etc. If they are inside the engine or engine nacelle . The cause of the window being blown out was that it had been refitted using undersized screws. The result was a cost of several million pounds sterling and a number of aircraft being out of service for considerably longer than they should have been. A few years ago. An airline pilot had a very lucky escape when he was nearly sucked through a window in the cockpit. a team of “experienced” mechanics thought they knew how to do a particular maintenance task so did not follow the instructions in the maintenance manuals. These are extreme examples of what may be considered as “maintenance induced failures”. A few examples of maintenanceinduced failure are discussed in this section. The rapid loss of pressure caused the pilot sat next to the window to be sucked through the hole. Although this was not entirely the fault of the refuellers. A number of recommendations followed this incident which should mean that it never happens again (provided everyone follows the procedures correctly). One of the major causes for accidental damage to components (from line replaceable units to parts) is the need to remove them in order to access other components. Similarly.
Repair and maintenance of building stock in the UK represents over 5% of Gross Domestic Product. serious damage. MAINTENANCE COST The world's airlines spend around $21 billion on maintenance. We have already seen that there are many factors which can affect the costs of maintaining a system. However. 31% on engine overhaul. 39. this expenditure is primarily the result of poor quality and unreliability. until practical electronic tagging of all parts becomes available. safety and integrity. over the life of many types of buildings.5. since it is impossible to produce a system which will never fail if operated for long enough we must consider ways in which the costs of maintenance can be kept to a minimum whilst ensuring system availability. out of which 21% is spent on line maintenance.1996]. which may again have serious consequences. There may be a temptation to use alternative sources (other than those authorised). possible. Configuration control and full traceability of parts is an essential element of aircraft safety but. it will remain difficult to police effectively. . In many cases these may be made from inferior materials or to less demanding tolerances and quality standards. 16% on component overhaul and the remainder on modifications and conversions (M Lam 1995). nonetheless. 27% on heavy maintenance. or £36 billion at 1996 [Building maintenance information report 254. Some spare parts may be expensive or difficult to obtain. the operators and maintainers of the system can. Consistent and sensible use of fasteners can not only reduce such problems but will also reduce the parts list and hence improve the supportability of the aircraft. If one recognises that maintenance is essentially the management of failure then clearly. The use of such rogue parts may result in premature component failure and. Fasteners overtightened may cause distortion resulting in leaks or damage. Maintenance and repair costs can be two to three times the initial capital costs. do much to minimise the cost of ownership by adopting the most suitable maintenance policies for the conditions prevailing. Whilst the original design will be a major influencing factor on these costs. Maintainability and Maintenance 177 they may be sucked into the delicate machinery almost certainly causing extensive and expensive damage.
C f = cost of facilities and C d = cost of technical data. facilities. C m = cost of material. which consist of all costs other than direct material. the total maintenance cost throughout the life of a systems/product is the sum of the corrective and preventive maintenance costs and the overhead costs.3 Indirect cost of maintenance task Indirect costs includes as management and administration staff needed for the successful completion of the task and the cost of the consequences of not having the system available which is related to a complete or partial loss of production and/or revenue. electricity. which is defined as: CMT = C s + C m + C p + C te + C f + C d (5. Similarly. 39. This is the cost of the maintenance resources directly used during the execution of the maintenance task. These costs should not be neglected.2 Direct cost of maintenance task The direct cost associated with each maintenance task. training and similar which are incurred while the item is in state of failure (and. labour and plant equipment. the expected preventive maintenance cost is the total cost of maintenance resources needed to inspect and/or examine an item before failure takes place and to replace any items rejected. taxes. The cost of maintenance task can be divided into two categories: 39. Maintainability and Maintenance 178 39. Thus.e. not included in the direct costs). . It also includes the overhead costs. heating. IT. telephone. The expected corrective maintenance cost is the total cost of maintenance resources needed to repair or replace failed items. because they could be even higher than the other cost elements. salaries of employers. insurance.5. of course. is related to the cost of maintenance resources. CMR. i. which are mentioned in Section 9. C te = cost of tools and support equipment. whether timebased or conditionbased.11) Where: C s = cost of spare parts.1 Cost of Maintenance Task The cost of the maintenance task is the cost associated with each corrective or preventive task. C p = cost of personnel. CMT.
2. is directly proportional to the product of the length of the time which the system spends in the state of failure (down time) and the income hourly rate. IHR. due to: 1. thus: CMT = CMR + CLR (5. if an airliner is not permitted to fly between the hours of 21:00 and 07:00 then any maintenance tasks undertaken and completed during those 10 hours would not affect the revenueearning capacity of the aircraft. 39. which is the money the system would earn whilst in operation. planned or scheduled maintenance would normally be done when the system would be expected to be idle and would only count as “downtime” for any period that the system would be expected to be operational. 4. 3. DMT I and DMT E .12) Where DMT is duration of maintenance task. preventative. DST is duration of support task and DT is total down time. DMT p .5.14) It is necessary to underline that the cost defined by the above expression could differ considerably. Note for systems that are not normally in continuous operation. Adoption of different maintenance policies The direct cost of each maintenance task Consumption of maintenance resources Duration of maintenance task. Thus.13) Making use of the above equations the expression for the cost of the completion of each maintenance task is defined as: CMT = C s + C m + C p + Cte + C f + Cd + ( DMT + DST ) × IHR (5. CLR. In particular. for example. DMT c . the downtime should take account of the proportion of the time the system would normally be expected to be operational. Maintainability and Maintenance 179 Cost of lost production and/or revenue. the cost of lost revenue could be determined according to the following expression: CLR = ( DMT + DST ) × IHR = DT × IHR (5.4 Total cost of maintenance task The total cost of maintenance task is the sum cost of direct and indirect costs. Thus.
DST c . Duration of support task.15) . DST I and DST E 7. the general expression for the cost of each maintenance task will have different data input for different maintenance policies. Indirect costs of maintenance tasks. Frequency of preventive maintenance task. Maintainability and Maintenance 180 5. FMT L . 9.5. DST I and DST E can take. NMT (Tst ) . DST p . NMT (Tst ) = 8. FMT I and frequency of examination FMT E 6. Different probability distributions and different values which random variables Tst MTTF DMT c . DMT I . CMT p is cost in the case of time based maintenance CMT I is cost of inspection based maintenance and CMT I is cost of examination based maintenance. the frequency of inspection. The expected total maintenance cost for a stated time. For example. as shown below: c c c c CMT c = C sp + C m + C c + C te + C c + C d + ( DMT c + DST c ) × IHR c p f p p p p CMT p = C sp + C m + C p + C te + C fp + C dp + ( DMT p + DST p ) × IHR p I I I I CMT I = C sp + C m + C p + C te + C If + C dI + ( DMT I + DST I ) × IHR I E E E E CMT E = C sp + C m + C p + C te + C E + C dE + ( DMT E + DST E ) × IHR E f Where: CMT c is related to the cost of each maintenance task performed after the failure. The expected number of maintenance tasks NMT (Tst ) performed during the stated operational length. in the case of FBM. thus: CMT (Tst ) = CMT c × NMT c (Tst ) + CMT p × NMT p (Tst ) + CMT I × NMT I (Tst ) + CMT E × NMT E (Tst ) (5. Thus. DST p . CMT (Tst ) . DST c . is equal to the product of the maintenance cost for each maintenance task and the expected number of maintenance tasks performed during the stated time. DMT E . DMT p . Lst .
Federal Aviation Regulation (FAR) require that. and so on. Each airline develops its own maintenance plan. and so on. Maintainability and Maintenance 181 39. 2. Maintenance facility availability and utilisation. 3. and personnel may be used for other tasks. All aircraft must follow a maintenance program that is approved by a regulatory authority such as FAA (Federal Aviation Administration. 5. maintenance is one of the most important functions to assure safe operation. Life expectancy of system 8. Test and support equipment effectiveness. Thus eliminating all maintenance tasks at first line (or OLevel) may not necessarily lead to a significant reduction in the personnel deployed or. it is important to recognise that facilities. mechanics in the armed forces may be put on guard duty or provide a defence role when not performing maintenance tasks. AIRCRAFT MAINTENANCE . inspection intervals and related procedures or alternative inspection intervals and related procedures set forth in the operations specifications or inspection program has been complied with. based on the manufacturer’s recommendations and by . equipment. Supply responsiveness or the probability of having a spare part available when needed. Maintenance task resources In order to reduce maintenance costs. Maintenance organisational effectiveness and personnel efficiency. UK). levels of inventory. which is the reliability and availability of test equipment. Durability and reliability of items in the system 7. supply lead times for given items. In calculating the various cost elements of maintenance.5 Factor Affecting Maintenance Costs Maintenance cost could be affected by the following factors: 1. For example. 40. test equipment utilisation. 6. it is necessary that the impact of the above factors should be reduced and/or controlled. 4.5. system test thoroughness.CASE STUDY For every commercial airline. in the operational costs. Duration of maintenance and support task 10. indeed. USA) and CAA (Civil Aviation Authority. Transportation times between maintenance facilities. no person may operate an aircraft unless the mandatory replacement times. Expected number of maintenance tasks 9.
5. Maintainability and Maintenance
182
considering its own operation. Thus, two different airlines may have slightly different maintenance program for same aircraft model used under similar operating conditions. Aircraft maintenance is reliability centred. It is claimed that each aircraft receives approximately 14 hours of maintenance for every hour it flies (R Baker, 1995). Maintenance accounts for approximately 10% of an airline’s total costs. On average a typical Boeing 747 will generate a total aircraft maintenance cost of approximately $1,700 per block hour. Aircraft maintenance can be categorised as: 1. 2. 3. 4. Routine scheduled maintenance. Nonroutine maintenance. Refurbishment. Modifications.
Routine Scheduled Maintenance Scheduled maintenance tasks are required at determinant recurring intervals or due to Airworthiness Directives (AD). The most common routine maintenance is visual inspection of the aircraft prior to a scheduled departure (known as walk around) by pilots and mechanics to ensure that there are no obvious problems. Routine maintenance can be classified as: 1. Overnight maintenance. 2. Hard time maintenance. 3. Progressive Inspection. Overnight maintenance normally includes low level maintenance checks, minor servicing and special inspections done at the end of the working for about one to two hours to ensure that the plane is operating in accordance with Minimum Equipment List. Overnight maintenance provides an opportunity to remedy passenger and crew complaints (M Lam, 1995). Hard time is the oldest primary maintenance process. Hard time requires periodic overhaul or replacement of affected systems/components and structures and is flight, cycle and calendar limited. That is, as soon as the
5. Maintainability and Maintenance
183
component age reaches it hard time it is replaced with a new component. Most of the rotating engine units are hard timed. The purpose of hard time maintenance is to assure operating safety of component or system, which have a limited redundancy. Progressive inspection groups like time related maintenance tasks into convenient ‘blocks’ so that maintenance workload becomes balanced with time and maintenance can be accomplished in small ‘bites’ making equipment more available. Grouping maintenance tasks also helps better utilisation of the maintenance facilities. These maintenance task groups are (detailed information can be found in M Lam (1995) and L R Crawford, 1995): 1. Preflight Visual inspections carried out by the mechanic and the pilots to ensure that there are no obvious problems. 2. A Check Carried out approximately every 150 flight hours, which includes selected operational checks (general inspection of the interior/exterior of the aircraft), fluid servicing, extended visual inspection of fuselage exterior, power supply and certain operational tasks. During A check, the aircraft is on ground for approximately 8 to 10 hours and requires approximately 60 labour hours. 3. B Check Occurs about every 750 flight hours and includes some preventive maintenance such as engine oil spectroanalysis, oilfilter are removed and checked, lubrication of parts as required and examination of airframe. Also incorporates Acheck. The aircraft could be on ground for 10 hours and will require approximately 200 labour hours. 4. C Check Occurs every 3, 000 flight hour (approximately 15 months) and includes detailed inspection of airframe, engines, and accessories. In addition, components are repaired, flight controls are calibrated, and major internal mechanisms are tested. Functional and operational checks are also performed during Ccheck. It also includes both A and B checks. The aircraft will be on ground for 72 hours and will require approximately 3,000 labour hours. 5. D Check This is the most intensive form of routine maintenance occurs about 20,000 flight hours (six to eight years). It is an overhaul that returns the aircraft to its original condition, as far as possible. Cabin interiors including seats, galleys, furnishings etc are removed to allow careful structural inspections. The aircraft is on ground for about 30 days and will require approximately 20,000 labour hours. A and B checks and overnight maintenance are instances of line maintenance (performed upon the aircraft incidental to its scheduled revenue operations), often carried out an airport. C and D checks, however
5. Maintainability and Maintenance
184
are heavy maintenance that requires special facilities and extensive labour. The task intervals for various checks mentioned above could vary significantly. The recommended time intervals for different aircraft models are given in Table 5.3 (Aircraft Economics). Table 5.3 Different scheduled checks in a commercial aircraft Aircraft Type Boeing 707 Boeing 727 Boeing 737100 Boeing 747100 DC8 DC9 A check Flight hours 90 80 125 300 150 130 540 680 400 750 B Check C Check 450 1,600 3,000 3,600 3,325 3,380 D Check 14,000 16,000 20,000 25,000 23,745 12,600 Flight hours Flight hours Flight hours
Nonroutine maintenance refers to the maintenance tasks that has to be performed on regular basis during checks, but which is not specified as routine maintenance task on the job cards of the maintenance schedule. Nonroutine maintenance shouldn’t be confused with unscheduled maintenance, which is repairs that have to be done as a result of an unexpected failure such as accidental damage (such as bird strike) to critical components or a response to airworthiness directives (AD). As the aircraft age, they require more maintenance due to fatigue and corrosion. The most significant of these aging aircraft airworthiness directives concerns Boeing 747. The fuselage of the Boeing 747 is built in sections as separate entities and then assembled during the aircraft production phase. The fuselage is built in five sections and the points at which these sections are joined are called the production breaks. Section 41 is the section from the nose to just aft of the forward passenger entry (Maintaining the Boeing 747, Aircraft Economics, 1994). The modification of Section 41, which is the area ahead of the forward passenger doors, requires approximately 60,00070,000 manhours to complete and requires replacement of most of the structural components (L Crawford, 1995).
5. Maintainability and Maintenance
185
5. Maintainability and Maintenance
186
Availability, Design for Reliability and Data Analysis
Course Material
Course Instructor: Professor U Dinesh Kumar Indian Institute of Management Bangalore
10. Any delay may cause it to miss its take off slot and more significantly its landing slot. is not beneficial to its owner. As mentioned in Chapter 9. missing its slot could mean a delay of several hours. in fact. Designers and manufacturers know that they are unlikely to remain in business for very long if their systems do not satisfy the customers’ requirements in terms of operational effectiveness. The whole operation generally takes about an hour. It may disrupt the timetables and cause problems for several days. If the particular flight was due to land just before the airport closes. clean the cabin. operational availability and achieved availability are some of the measures used to . Inherent availability. restock with the next flight’s foods and other items. but inconsistency Jonathan Swift Availability is used to measure the combined effect of reliability. Many forms of availability are used to measure the effectiveness of the system. which is in a state of failure. refuel. This is likely to cause inconvenience to the customers who may then decide to switch to an alternative airline in future. Many airports close during the night to avoid unacceptable levels of noise pollution. and reload with the next set of passengers and baggage. it is probably costing the owner money. maintenance and logistic support on the operational effectiveness of the system. Availability 187 Chapter 10 Availability There is nothing in this world constant. A system. it cannot be used until it has been declared airworthy. even though this may be 12 hours later. If an aircraft breaks down. since an aircraft cannot takeoff until it has been cleared to land. most large airliners have a very high utilisation rate with the only down time being to do a transit check. unload. An operator of a system would like to make sure that the system will be in a state of functioning (SoFu) when it is required.
consider an item with constant failure rate λ and constant repair rate µ. For example. Reliability Maintainability Supportability Availability Figure 10.1). renewal process. 10. Availability is defined as: The probability that an item is in state of functioning at a given point in time (point availability) or over a stated period of time (interval availability) when operated.1 Availability as a function of reliability. the item can be in either the state of functioning (say. maintainability and supportability factors (Figure 10. operational availability and achieved availability. regenerative process. maintained and supported in a prescribed manner. we look at few important availability measures such as point availability. maintainability and supportability In this chapter. POINT AVAILABILITY Point availability is defined as the probability that the system is in the state of functioning (SoFu) at the given instant of time t. At any instant of time.41. Availability 188 quantify whether an item is in an operable state when required. state 1) or in the state of failure (say. We use the notation A(t) to represent the point availability.10. As both failure and repair . interval availability. Depending on the time to failure and time to repair distributions. semiMarkov process and semiregenerative process models to derive the expression for point availability. It is clear from the above definition that availability is a function of reliability. steady state inherent availability. one can use Markov chain. Availability expressions for systems can be obtained by using stochastic processes. state 2).
the expression for P2(t+h) can be written as: (10. Pi(t+h). The corresponding expression is given by p 21 ( h) = 1 − exp( −µh) ≈ µh for hµ<<1 p12(h) is the probability of entering state 2 from state 1 during the interval h.1) P2 (t + h) = P1 (t ) × p12 (h ) + P2 (t ) × p 22 (h ) (10.10. 2. The system was in state 2 at time t and it transits to state 1 during the interval h. The system was in state 1 at time t and continues to remain in state 1 throughout the interval h. The corresponding expression can be written as: P1 (t + h) = P1 (t ) × p11 (h ) + P2 (t ) × p 21 (t ) Using similar logic. as the probability that the system would be in state i at time t+h. Availability 189 rates are constant (and thus follow exponential distribution). 2. for i = 1. Let pij(h) denote the transition probability from state i to state j during the interval ‘h’(i. 2). The expression for P1(t+h) can be derived using the following logic: 1. Define. The probability p22(h) is given by: p 22 (h) = exp( −µh) ≈ 1 − µh for hµ<<1 . we can use a Markov chain to model the system to derive the availability expression. The probability p12(h) is given by p12 ( h) = 1 − exp( −λh ) ≈ λh for hλ<<1 p22(h) is the probability of remaining in state 2 during the interval h. The probability p11(h) is given by p11 (h) = exp( −λh) ≈ 1 − λh for λh<<1 p21(h) is the probability of entering state 1 from state 2 during the interval h.2) p11(h) is the probability of remaining in state 1 during the interval h.j = 1.
we have Lt P1 (t + h) − P1 (t ) dP1 (t ) = = −λP1 (t ) + µP2 (t ) h dt P2 (t + h) − P2 (t ) dP2 (t ) = = λP1 (t ) − µP2 (t ) h dt h →0 h →0 Lt On solving the above two differential equations. the point availability A(t) is given by: A(t ) = µ λ + × exp( −(λ + µ )t ) λ+µ λ+µ (10.10.1) and (10. then the point availability A(t) can be written as (Birolini. we get 190 P1 (t + h ) = P1 (t ) × (1 − λh) + P2 (t ) × µh P2 (t + h) = P1 (t ) × (λh) + P2 (t ) × (1 − µh) By rearranging the terms and setting h → 0.4) × exp( −( + MTTF + MTTR MTTF + MTTR MTTF MTTR When the time to failure and time to repair are not exponential. we can use a regenerative process to derive the availability expression. 1997): A(t ) = 1 − F (t ) + ∫ ∑ [ f ( x) * g ( x )]n [1 − F (t − x)]dx 0 n =1 t ∞ . we get P1 (t ) = µ λ + × exp( −(λ + µ )t ) λ+µ λ+µ P1(t) is nothing but the availability of the item at time t.3) Substituting λ = 1/MTTF and µ = 1/MTTR in the above equation. Availability Substituting the values of pij(h) in equation (10. Thus. we get A(t ) = 1 1 MTTF MTTR + )t ) (10. that is the probability that the item will be in state of functioning at time t.2). If f(t) and g(t) represent the timetofailure and timetorepair distributions respectively.
t] that the system is in state of functioning. is defined as the expected fractional duration of an interval (0. Thus. t]. Ai.10. the following inequality is true.7) The above result is valid for any time to failure function F(t) and any time to repair distribution G(t) (Birolini.5) where A(x) is the point availability of the item as defined in equation (10. Also. assuming that this probability depends only on the timetofailure and time to repair distributions.4). . in the case of constant failure rate λ and constant repair rate µ.1 Average Availability Interval availability. and 1 F(tx) is the probability that no failures occur in the remaining interval [x. Thus. The summation n =1 n ∑ [ f ( x) ∗ g ( x )] gives the renewal points f(x)*g(x). is defined as the steady state probability (that is.6) 10.41. t → ∞) that an item will be in a state of functioning. 1997). For an item with constant failure rate λ and constant repair rate µ. Availability 191 where [f(x)*g(x)]n is the nfold convolution of f(x)*g(x). x+dx].3) and (10. the inherent availability is given by: Ai = Lt A(t ) = t →∞ MTTF MTTF + MTTR (10. ∞ lies in [x. It is assumed that any support resources that are required are available without any restriction.. 10. AA(t). the average availability is given by: AA(t ) = µ λ + [1 − exp( −(λ + µ )t )] λ + µ (λ + µ ) 2 t (10.2 Inherent Availability Inherent availability (or steadystate availability). f(x)*g(x)*f(x)*g(x). 1t AA(t ) = ∫ A( x) dx t0 (10.41.
Plot the point availability of the DECU. The average availability of the system during 5000 hours of operation is given by: AA(t ) = µ λ + [1 − exp( −(λ + µ )t )] λ + µ (λ + µ ) 2 t Substituting the values of λ (= 1/1200) and µ (=1/400).2 0 0 1000 2000 3000 Tim e 4000 5000 6000 Steadystate Figure 10. 1 Ava ila bility 0. 1.2 depicts the point availability of the system. .4 0. 2. The point availability of the DECU is calculated using the equation (10.6 0.4). Availability A(t ) − Ai ≤ exp( −t / MTTR) 192 (10.8 0. 3. SOLUTION: 1. Figure 10. Find the average availability of the DECU during first 5000 hours. Find the inherent availability.1 Time to failure distribution of a digital engine control unit (DECU) follows an exponential distribution with mean time between failures 1200 hours and the repair time also follows an exponential distribution with mean time to repair 400 hours. we get the value of the average availability during 5000 hours as 0.7649.2 Point availability of DECU 2.8) Example 10.10.
75 MTTF + MTTR 1200 + 400 193 Thus. The inherent availability of the system is given by Ai .10) For a series system with all the elements having constant failure and repair rates.9) where Ai(t) is the point availability of ith item.11) MTTFs and MTTRs are system mean time to failure and system mean time to repair respectively.3 System Availability of different reliability block diagrams Availability of a system with series reliability block diagram with n items is given by As (t ) = ∏ Ai (t ) k =1 n (10. Availability 3. the system inherent availability Ai . where λs = ∑ λi λs i =1 i =1 . Let λi and µi represent the failure rate and repair rate of item i respectively. MTTFs and MTTRs are given by MTTFs = 1 i =1 n ∑ λi n MTTRs = ∑ n λi MTTRi .41.s = MTTFs MTTFs + MTTR s (10.75 or 75%.10. the steady state availability of the system is 0. The inherent availability is given by Ai = MTTF 1200 = = 0.s = ∏ MTTFi k =1 MTTFi + MTTRi n (10. 10.
Find the inherent availability of the system. The time to failure and the time to repair distributions of the different items are given as given in Tables 10.1 and 10.7 µ = 800 hours σ = 180 hours Item Number Item 1 Item 2 Item 3 Item 4 Distribution Weibull Exponential Weibull Normal Table 10.75 µ = 72 hours σ = 24 hours Item number Item 1 Item 2 Item 3 Item 4 SOLUTION: Distribution Lognormal Normal Lognormal Normal First we calculate MTTFi and MTTRi for different items: .25 µ = 48 hours σ = 12 hours µl = 3.5 and σl = 0.25 and σl = 1.2 A series system consists of four items.7 λ = 0.10.1.12) Example 10. Time to failure distribution for different items.2.0008 per hour η = 1800 hours β = 2. Table 10. Availability Availability of a parallel system with n items is given by As (t ) = 1 − ∏ [1 − Ai (t )] i =1 n 194 (10.2. Parameters η = 2200 hours β = 3. Time to repair distribution for different items Parameters µl = 3.
8362 i =1 4 10. A2 = 0.0008 = 1250 . Tsm is the scheduled maintenance interval (time between scheduled maintenance). Substituting the values of MTTFi and MTTRi in equation (10. MTTF3 = 1600.13) MTBM is the mean time between maintenance and AMT is active maintenance time.11). T.7 MTTF2 = 1 / λ = 1 / 0. ACHIEVED AVAILABILITY Achieved availability is the probability that an item will be in a state of functioning (SoFu) when used as specified taking into account the scheduled and unscheduled maintenance.10. Achieved availability. Ai.2 . is given by Aa = MTBM MTBM + AMT (10.9174 The system availability is given by As = ∏ Ai = 0. the item is ‘asbadasold’ and after each . The mean time between maintenance during the total operational life.33 hours. for item i can be calculated using the equation (10. MTTR4 = 72 hours Inherent availability. we have A1 = 0.87 hours. Availability MTTF1 = η × Γ(1 + 195 1 1 ) = 2200 × Γ(1 + ) = 2200 × 0. The above expression is valid when after each scheduled maintenance. A4 = 0. A3 = 0.4 β 3. MTTR2 =48 hours MTTR3 = exp( µ l + σ l 2 / 2) = 43. that is the expected number of failures during the total life T. Aa.42. is given by: MTBM = T M (T ) + T / Tsm (10.14) M(T) is the renewal function.9630.11). any support resources needed are available instantaneously.9733. MTTF4 =800 MTTR1 = exp( µ l + σ l 2 / 2) = 56.9723.902 = 1984.
3 Time to failure distribution of an engine monitoring system follows a normal distribution with mean 4200 hours and standard deviation 420 hours. MTBM.1434 MTBM = 20000 ≈ 1414 hours 4. is given by MTBM = T 20000 = M (T ) + T / Tsm M ( 20000) + 20000 / 2000 M(20000) for normal distribution with mean 4200 hours and standard deviation 420 hours is given by M (20000) = ∑ Φ ( n =1 ∞ 20000 − n × 4200 n × 420 ) = 4.000 hours (subject to corrective and preventive maintenance). A scheduled maintenance is carried out after every 2000 hours and takes about 72 hours to complete the task.10. The engine monitoring system is expected to last 20. is given by: AMT = M (T ) × MTTR + (T / Tsm ) MSMT M (T ) + T / Tsm 196 The active (10. AMT. Example 10. Availability corrective maintenance the item is ‘asgoodasnew’. maintenance time. SOLUTION: Mean time between maintenance.15) MTTR stands for the mean time to repair and MSMT is the mean scheduled maintenance time. The time to repair the item follows a lognormal distribution with mean time to repair 120 hours.1434 + 10 The active maintenance time is given by: . Find the achieved availability for this system.
T.1434 × 120 + 10 × 72 ≈ 86. is given by Ao = MTBM MTBM + DT (10. The system down time DT is given by: DT = M (T ) × MTTRS + (T / Tsm ) MSMT M (T ) + T / Tsm (10. MTBM is the mean time between maintenance (including both scheduled and unscheduled maintenance) and DT is the Down time.06 4.18) MTTRS stands for the mean time to restore the system and MSMT is the mean scheduled maintenance time. that is the expected number of failures during the total life T. MTTRS is given by . Ao. The mean time between maintenance during the total operational life.9426 MTBM + AMT 1414 + 86.16) where.43.1434 + 10 197 The achieved availability of the system is given by: Aa = MTBM 1414 = = 0. is given by: MTBM = T M (T ) + T / Tsm (10.10. OPERATIONAL AVAILABILITY Operational availability is the probability that the system will be in the state of functioning (SoFu) when used as specified taking into account maintenance and logistic delay times. Operational availability. Availability AMT = = M (T ) × MTTR + (T / Tsm ) MSMT M (T ) + T / Tsm 4.06 10.17) M(T) is the renewal function. Tsm is the scheduled maintenance interval (time between scheduled maintenance).
assume that whenever a system fails it takes about 48 hours before all the necessary support resources are available.19) Example 10. Find the operational availability.1434 × 168 + 10 × 72 = 100. In the absence of any scheduled maintenance the operational availability can be calculated using the following simple formula AO = MTBF MTBF + MTTR + MLDT (10. Availability MTTRS = MTTR + MLDT 198 where MLDT is the mean logistic delay time for supply resources. SOLUTION MTBM is same as in the previous example and is equal to 1414 hours.12 hours 14. The mean time to restore the system is given by MTTRS = MTTR + MLDT = 120 + 48 = 168 hours The system down time is given by DT = M (T ) × MTTRS + (T / Tsm ) MSMT M (T ) + T / Tsm = 4.12 .10.9338 MTBM + DT 1414 + 100.4 In the previous example.1434 The operational availability of the system is given by AO = MTBM 1414 = = 0.
Maintenance and Logistic Support Reliability. RELIABILITY ALLOCATION Reliability allocation is a process by which the system’s reliability requirements is divided into subsystem and component reliability requirements. Design phase is particularly important for any product as the decisions made during this stage can determine how reliable the product is going to be as well as the maintainability and supportability of that product. EFFECTS AND CRITICALITY ANALYSIS (FMECA) The failure modes. FAILURE MODES. 45. maintenance and supportability should be designed into the product. we would like to discuss few tool and techniques that can be used at the design stage to improve the RMS characteristics. Design for Reliability. . In this chapter. Maintenance and Logistic Support 199 Chapter 11 Design for Reliability. 44. effects and criticality analysis (FMECA) is a systematic method for examining all modes through which a failure can occur.11.
The criticality analysis examines how critical a failure would be for the operation and safe use of the product. The criticality might range from minor failure through lowering of performance. The FMECA concept was developed by US defence industries in the 1950s. 1993).11. maintenance. and familiarity with the product/process being analysed and its design and functionality (D Verma. location etc. extend of damage. Input requirements for FMECA analysis include reliability data. safety and supportability problems resulting from the effects of a product/process failure. Failure mode analysis lists all possible mode the failure would occur which include the condition. The sequence of steps followed to . their modes of failure. and the estimated criticality of the failures. to improve the reliability of military equipment. The failure effect analysis includes the study of the likely impact of failure on the performance of the whole product and the process. the probabilities of detection for the various failure modes are also required. Additionally. Maintenance and Logistic Support 200 potential effects of these failures on the system performance and their relative severity in terms of safety. It is an excellent methodology for identifying and investigating potential product weaknesses. This analysis is best utilised during the early design and development phase of new systems. FMECA has become an important tools applied by almost all industries around the world to improve the reliability. It is claimed that a more rigorous FMECA analysis would have avoided the disastrous explosion of the Challenger launch on 28th January 1986. and impact on mission success. The three principal study areas in FMECA analysis are the failure mode. FMECA establishes a detailed study of the product design. maintainability and supportability of their product. 45. Since then.1 Procedural Steps in the FMECA analysis The procedural steps in FMECA analysis depend to a certain extend what product or process is being examined. failure effect and failure criticality. safety and environmental hazard to a catastrophic failure. manufacturing operation or distribution to determine which features are critical to various modes of failure. the components involved. and in the evaluation of existing system (D Verma. shutdown of the product. 1993). The actual FMECA performed could be both quantitative and qualitative based on the information available to the analyst. FMECA is performed to identify reliability. A prerequisite for the successful completion of FMECA is good knowledge of. Design for Reliability.
5. Maintenance and Logistic Support 201 accomplish the failure modes. Determine cause of failures. modules and components. Accomplish requirements allocation. Failure criticality can be classified in any one of four categories. 8. transportation and handling. Identify the criticality of failure. maintenance induced failures. 3. 6. which could be design and manufacturing deficiency. Design for Reliability. Effect of failure might range from catastrophic failure to minor performance degradation. Identification of all possible failure modes for the system as well as the subsystem. The following are the key steps involved in the FMECA analysis: 1. which employs functional approach as a basis for identification of design requirements for each hierarchical level of the system. Functional analysis is accomplished through functional flow diagram that portrays the system design requirements illustrating series and parallel relationships and functional interfaces). accidental damage. Accomplish functional analysis (Functional analysis is a systematic approach to system design and development. What is expected from the system in terms of operation and performance. effect and criticality analysis is depicted in Figure 11. depending upon the failure effects as follows . that is for a specified requirement at system level. 2. Assess the probability of failure. Identification of the system requirements. by defining the basic requirements for the system in terms of input criteria for design. 4.1. What is the customer requirements with respect to reliability. During the system requirement definition. 7. This can be achieved by analysing the failure data and identifying the timetofailure distribution. what should be specified at unit and assembly level. maintainability and supportability specified at system level are allocated to unit and assembly level.11. Identify the effects of failure. ageing and wearout. the following tasks should be addressed (Refer to Blanchard and Fabrycky 1999 for detailed discussion). What are the requirements for disposal after the system is withdrawn from service. System effectiveness factors such as reliability. maintainability and supportability How the system is used in terms of hours of operation/number of cycles per day etc.
10. Design for Reliability. b) Major failure Any failure that will degrade the system performance beyond an acceptable limit. the severity of the effects and the likelihood of detecting a failure mode. Define System Requirements Accomplish Functional Analysis Requirements Allocation Identify Failure Modes Determine Causes of Failure Determine Effects of Failure Assess the probability of Failure Assess the probability of Failure Assess the Criticality of Failure Feedback and corrective action loop Analyse failure mode Criticality (PRN) Figure 11. Compute the Risk Priority Number (RPN) by multiplying the probability of failure. Initiate corrective action that will minimise the probability of failure or effect of failure that show high RPN.11. 9.1 Sequence of steps involved in FMECA . d) Catastrophic failure Any failure that could result in significant system damage and may cause damage to property. Maintenance and Logistic Support 202 a) Minor failure Any failure that doesn t have any noticeable affect on the performance of the system. serious injury or death. c) Critical failure Any failure that would affect safety and degrade the system beyond an acceptable limit.
Rating scales for occurrence of failure Description Rating Remote probability of occurrence Low probability of occurrence Moderate probability of occurrence High probability of occurrence Very High probability of occurrence 1 23 46 7 8 9 .3 gives possible ratings for probability of failure. Tables 11.3. Maintenance and Logistic Support 203 45.1 – 11. As mentioned earlier. Note that. FS is the failure severity and FD denotes the failure detection probability. FP is the Failure probability. Table 11. severity of failure and failure detection. Design for Reliability.3 are only suggested ratings.11.1) Where.111.2 Rating scales for severity of failure Description Rating Minor failure Major Failure Critical Failure Catastrophic Failure 1 2 3 5 6 9 10 .10 Table 11. RPN is calculated by multiplying the probability of failure. That is: RPN = FP × FS × FD (11. the severity of the effects of failure and likelihood of failure detection.2 Risk Priority Number Risk Priority Numbers play a crucial role in selecting the most significant item that will minimise the failure or effect of failure. the ratings given in the tables 11.
06 0.00 Assume that a failure mode has following ratings for probability of failure.36 0.76 1. system safety hazard analysis. This is usually achieved using Pareto analysis with a focus on failure mode.75 9 . Outputs from a properly conducted FMECA can be used in developing a cost effective maintenance analysis. Maintenance and Logistic Support Description Rating 204 Table 11. .16 0.35 68 0. Design for Reliability. Rating scales for detection of failure Description Remote probability of detection Low probability of detection Moderate probability of detection High probability of detection Very high probability of detection Rating 1 Probability of Detection 0 0. failure cause and failure criticality. and logistic support analysis. Risk priority number for all the failure modes are calculated and priority is given to the one with highest RPN for eliminating the failure.05 23 0.15 45 0.11.3.10 0. failure severity and failure detection: Failure probability = 7 Failure severity = 4 Failure detection = 5 Then the risk priority number for this particular failure mode is given by 7 × 4 × 5 = 140.
1. One of the outputs from a fault tree analysis is the probability of occurrence of the toplevel event or failure. During the very early stages of system design process. It is necessary to specific in defining the toplevel event.11. Maintenance and Logistic Support 205 46. Usually. a separate fault tree is developed for every critical failure mode or undesired TopLevel event. a standard symbol is used to develop fault trees.4 depicts the symbols used to represent the causal hierarchy and interconnects associated with a particular toplevel event. This ‘TopDown’ causal hierarchy and the associated probabilities. the next step is to construct the initial causal hierarchy in the form of a fault tree. While developing the fault tree all hidden failures must be considered and incorporated. FAULT TREE ANALYSIS (FTA) Fault tree analysis is a deductive approach involving graphical enumeration and analysis of the different ways in which a particular system failure can occur. The following steps are used to carry out FTA (Figure 11. and the probability of its occurrence. Identify the toplevel event The most important step is to identify and define the toplevel event. Design for Reliability. If this probability is unacceptable. The FTA can have most impact if initiated during the conceptual and preliminary design phase when design and configuration changes can be most easily and cost effectively implemented. a generic and nonspecific definition is likely to result in a broad based fault tree which might be lacking in focus. fault tree analysis provides the designers with an insight into aspects of the system to which redesign can be directed or compensatory provisions be provided such as redundancy. Techniques such as Ishikawa s cause and effect diagram can prove beneficial. It starts with a toplevel event (failure) and works backward to identify all the possible causes and therefore the origins of that failure. While . fault tree analysis (FTA) is often conducted to gain insight into critical aspects of selected design concepts. Table 11. Attention is focused on this toplevel event and the firsttier causes associated with it. The logic used in developing and analysing a fault tree has its foundations in Boolean Algebra. Develop the initial fault tree Once the toplevel event has been satisfactorily identified. and in the absence of information required to complete a FMECA. and this process is continued. is called a Fault Tree. Each firsttier cause is next investigated for its causes. For the sake of consistency.2). 2.
11. Determine the reliability of the toplevel event and 3. the output is realised only after all the associated inputs have been received. The rectangle represents an intermediate fault event. A circle represents the lowest level failure event. Maintenance and Logistic Support 206 constructing a fault tree it is important to break every branch down to a reasonable and consistent level of detail. A rectangle can appear anywhere in a fault tree except at the lowest level in the hierarchy. This symbol represents the AND logic gate. Identify Top Level Event Develop the Initial Fault Tree Analyse the Fault Tree Review Analysis Output Determine TopLevel Event Reliability Delineate the Minimal Cutsets Figure 11. Design for Reliability. undeveloped events have a substantial amount of complexity below and can be analysed through a separate fault tree.2 Steps involved in a fault tree analysis. In this case. 3. The important steps in completing the analysis of a fault tree are 1.4. 2. Very often. also called a basic event. Fault tree construction symbols Description Table 11. Review analysis output. . Symbol The Ellipse represents the toplevel event (thus always appears at the very top of the fault tree). The diamond represents an undeveloped event. Analyse the Fault Tree The third step in FTA is to analyse the initial fault tree developed. which can be further broken. Delineate the minimum cutsets.
D Verma. Drive/suspension assembly. Maintenance and Logistic Support Symbol Description 207 This symbol represents the OR logic gate. Design for Reliability. FAULT TREE ANALYSIS PASSENGER ELEVATOR CASE STUDY In this section we discuss a case study on fault tree analysis of a passenger elevator (Main source. Control assembly and 2. Drive Unit Level 3 Control Unit Level 2 Weight Level 1 Passenger Car Shaft Power Supply Level 0 Figure 11.11. Consider a passenger elevator depicted in Figure 11. which awaits an operator signal request to move the car to a certain level.3 Schematic diagram of a passenger elevator The control assembly consists of a microprocessor.3. 47. The control unit activates drive unit that moves the car to that level and opens the elevator . In this case. 1993). any one or more of the inputs need to be received for the output to be realised. All drive assembly failures are generalised as ‘motor failures’ and ‘other failures’ while control unit failures are generalised as ‘hardware failures’ and ‘software failures’ for the sake of simplicity. We consider two major assemblies for FTA 1.
Maintenance and Logistic Support 208 door once the car comes to a stop. Elevator operating properly. 1 G1 2 3 Figure 11. Car stops between levels. B. Car falls freely. In this case.4 Initial fault Tree In Figure 11. The Drive unit moves or stops the car only when prompted to do so by the control unit.11. The brake unit is designed to hod the car stationary when power is removed and to allow the motor shaft to turn when power is applied. 2 and 3 are as defined below: Event 1 – Passenger injury occurs Event 2 – Car free falls Event 3 – Door opens without car present. operating conditions ‘C’ and ‘D’ are of concern. C. . the toplevel event (passenger injury occurs) can be either due to car free fall or door opens without the car present. Drive/suspension assembly holds the car suspended within the shaft and moves it to the correct level as indicated by the control unit. Switches exist at each level and inside the car allowing the controller to know where the car is at any time. The initial fault tree is shown in Figure 11.4. G1 is represents the OR logic gate and the events 1. Thus. D.4. Car entry door opens in the absence of car. Design for Reliability. We define the toplevel event in this case is ‘passenger injury occurs’. The following are the possible system operating conditions: A.
5. which can be further analysed using a separate fault tree. Maintenance and Logistic Support 209 Now the event. Design for Reliability. Event 5 is an intermediate event. 5 and 6 are defined below: Event 4 – Cable slips of pulley Event 5 – Holding brake failure Event 6 – Broken cable Event 4 and 6 are undeveloped event. In Figure 11.11.5. 2 G2 4 5 6 Figure 11.5 Further FTA analysis of the event car free fall . can further analysed by treating it as a toplevel event. car free fall. G2 is again a OR gate and the events 4. resulting in a fault tree depicted in Figure 11. which can further broken.
it is not covered in this book. Combining fault trees depicted in Figures 11. Figure 11. 8 and 9 can be further expanded to find the causes using fault tree analysis. Maintenance and Logistic Support 210 3 G4 7 8 G5 9 Controller Error Figure 11. passenger injury.7. As it is a simple mathematical calculation.6 depicts FTA for the event 3. If the derived toplevel probability is unacceptable. door opens without the car present. we can construct a complete (almost) for the event. events 7 and 8 must happen.411. necessary redesign or compensation efforts should be identified and initiated.6 Fault tree for the event.11. The probability for the occurrence of the toplevel event can be calculated once the timetofailure and probability of occurrence of all the events are known. door opens erroneously. to occur. This can be caused due to the following events: Event 7 – Door close failure Event 8 – Car not at level Event 9 – Latch failure For the event. 6. G4). Note that events 4.6. the door opens erroneously The event 3 can be further analysed to find the causes. The door close failure can be caused either due to the latch failure or due to controller error (denoted by OR gate. as shown in Figure 11. . 5. thus we have a AND gate G3. Design for Reliability.
7 Fault tree for the event.11. Design for Reliability. passenger injury . Maintenance and Logistic Support 211 1 G1 2 3 G2 G3 7 8 4 5 6 G4 9 Controller Error Figure 11.
Even relatively simple systems can fail in a number of different ways and for a number of different reasons. say). components will be replaced before they have actually failed. There is useful data to be gleaned from the ones that have not failed as well as from the ones that have failed. In the case of maintenance and supportability data. We may wish to determine whether a new version of a component is more reliable than a previous version to decide whether we have cured the problem (of premature failures. Thus. Probability papers. 1. We may have a number of systems undergoing testing to determine whether the product is likely to meet the various requirements but we need to go into production before they have all failed. In the section on “censored data” we recognise that very often we do not have a complete set of failure data. it is essential that we have sufficient information on the time to failure. time to repair (maintain) and time to support characteristics of that item. Linear regression. If we do not put the nut on squarely. If a component is being used in a number of different systems. Often. Suppose we wish to fasten two pieces of metal together using a nut and bolt. that is the data relating to the age of the components at the time they failed while they were in operation (in maintenance and logistic support we analyse the data corresponding to the maintenance and support task completion times). In most cases these characteristics are expressed using theoretical probability distributions. Maximum likelihood estimates to identify the best distribution using which the data can be expressed and to estimate the corresponding parameters of the distribution. they have been damaged or they are showing signs of excessive wear. we would identify the maintainability and supportability function as in the case of reliability data and then compute MTTR and MTTS. 2. as well as its maintainability and supportability function. then one can extract information about the type of the hazard function and other reliability characteristics such as mean time between failures and failure rate etc. Even though the way the different systems operate may be different. If we overtighten the nut. If the two pieces of metal are being forced apart then the stress on the nut and bolt may cause the thread to strip either inside the nut or on the outside of the bolt or it may cause the bolt to exceed its elastic and plastic limits until it eventually . To start with we look at ways of fitting probability distributions to inservice data. it is still likely that the shape of the failure distribution will be same and that only the scale will be different. it may be reasonable to assume that the failure mechanism in each of these instances will be similar.Chapter 12 Analysis of Reliability. Maintenance and Supportability Data Often statistics are used as a drunken man uses lamp posts… for support rather than illumination. we might strip the thread or we might shear the bolt. Once the distribution is identified. we could cross the threads and hence weaken the joint. We look at three popular tools. possibly because they have started to crack. the problem which every logistician face is the selection of the appropriate distribution function to describe the empirical data (obtained from data capturing sources) using theoretical probability distributions. and 3. To predict various reliability characteristics of an item.
breaks. If the joint is subject to excessive heat this could accelerate the process. Equally, if it is in very low temperatures then the bolt is likely to become more brittle and break under less stress than at normal temperatures. If the diameter of the bolt is towards the lower limit of its tolerance and the internal diameter of the nut is towards the upper limit then the amount of metal in contact may not be sufficient to take the strains imposed. As the two components age, corrosion may cause the amount of metal in contact to be even further reduced. It may also change the tensile strength of the metals and cause premature failure. Components may therefore fail due to a number of failure modes. Each of these modes may be more or less related to the age. One would not expect corrosion to be the cause of failure during the early stages of the component’s life, unless it was subjected to exceptionally corrosive chemicals. On the other hand, if the components have been badly made then one might expect to see them fail very soon after the unit has been assembled. Very often, a possibly small, number of components may fail unexpectedly early. On further investigation it may be found that they were all made at the same time, from the same ingot of metal or by a particular supplier. Such a phenomenon is commonly referred to as a batching problem. Unfortunately, in practice, although it may be possible to recognise its presence, it may not always be possible to trace its origin or, more poignantly, the other members of the same batch or, indeed, how many there may be. In deciding whether a new version of a component is more reliable than the old one, we need to determine how confident we are that the two distributions are different. If they both have the same (or nearly the same) shapes then it is a relatively straightforward task to determine if their scales are different. In some cases, the primary cause of failure of the origin version may have been eliminated or, at least, significantly improved but, another, hitherto rarely seen cause, may have become elevated in significance. This new primary cause may have a distinctly different shape than the first one that often makes it very difficult to decide between the two. In this chapter, we first look at the empirical approaches for finding estimates for MTTF, MTTR and MTTS as well as failure function, maintainability and supportability functions. Rest of the chapter describes some of the wellknown methods for selection of the most relevant theoretical distribution functions for the random variables under consideration.
12.48.
RELIABILITY, MAINTENANCE AND SUPPORTABILITY DATA
A very common problem in reliability engineering is the availability of failure data. In many cases getting sufficient data for extracting reliable information is the most difficult task. This may be due the fact that there is no good procedure employed by the operator (or supplier) to collect the data or the item may be highly reliable and the failure is very rare. However, even without any data, one should be able to predict the timetofailure distribution if not the parameters. For example, if the failure mechanism is corrosion, then it cannot be an exponential distribution. Similarly if the failure cause is ‘foreign object damage’ then the only distribution that can be used is exponential. The main problem with insufficient failure data is getting an accurate estimate for the shape parameter. Fortunately, we don’t have such problems with maintenance and supportability data. These are easily available from the people who maintain and support the item. The reliability data can be obtained from the following sources: 1. Field data and the inservice data from the operator using standard data capturing techniques. There are standard failure reporting forms for the purpose of capturing desired information regarding the reliability of the item under consideration. Unfortunately, all these forms are 213
2.
flawed, as they record only MTBF (or MTTR and MTTS in case of maintenance and support). Just the value of MTBF alone may not be enough for many analyses concerning reliability (similarly, in the case of maintenance (support), information on MTTR (MTTS) is not enough for complete analyses). From life testing that involves testing a representative sample of the item under controlled conditions in a laboratory to record the required data. Sometimes, this might involve ‘accelerated life testing’ (ALT) and ‘highly accelerated life testing’ (HALT) depending on the information required.
As mentioned earlier, in some cases it is not possible to get a complete failure data from a sample. This is because some of the items may not fail during the life testing (also in the inservice data). These types of data are called ‘censored data’. If the life testing experiment is stopped before all the items have failed, in which cases only the lower bound is known for the items that have not failed. Such type of data is known as ‘right censored data’. In few cases only the upper bound of the failure time may be known, such type of data is called ‘left censored data’.
12.49.
ESTIMATION OF PARAMETERS  EMPIRICAL APPROACH
The objective of empirical method is to estimate failure function, reliability function, hazard function, MTTF (or MTTR and MTTS) from the failure times (or repair and support times). Empirical approach is often referred as nonparametric approach or distribution free approach. In the following sections we discuss methods for estimating various performance measures used in reliability, maintenance and support from different types of data.
Estimation of Performance Measures  Complete Ungrouped Data
Complete ungrouped data refers to a raw data (failure, repair or support) without any censored data. That is, the failure times of the whole sample under consideration are available. For example, let t1, t2, …, tn, represents n ordered failure times such that ti ≤ ti+1. Then the possible estimate for failure function (cumulative failure distribution at time ti) is given by:
F (t ) =
^
i n
(12.1)
A total of i units fail by time t out of the total n in the sample. This will make F(tn) = n /n = 1. That is, there is a zero probability for any item to survive beyond time tn. This is very unlikely, as the times are drawn from a sample and it is extremely unlikely that any sample would include the longest survival time. Thus the equation (12.1) underestimates the component survival function. A number of mathematicians have tried to find a suitable alternative method of estimating the cumulative failure probability. These range from using n+1 in the denominator to using 0.5 in the numerator and +0.5 in the denominator. The one that gives the best approximation is based on median rank. Bernard's approximation to the median rank approach for cumulative failure probability is given by
F (t i ) =
^
i − 0. 3 n + 0. 4
(12.2)
214
Throughout this chapter we use the above approximation to estimate the cumulative failure distribution or failure function. From equation (12.2), the estimate for reliability function can be obtained as
R(ti ) = 1 − F (t i ) = 1 −
^ ^
i − 0.3 n − i + 0.7 = n + 0. 4 n + 0. 4
(12.3)
The estimate for the failure density function f(t) can be obtained using
^
F (t i ) − F (t i +1 ) f (t ) = , t i − t i +1
^
^
t i ≤ t ≤ t i +1
(12.4)
Estimate for the hazard function can be obtained by using the relation between the reliability function R(t) and the failure density function f(t). Therefore,
h(t ) = f (t ) R(t ) for t i < t < t i +1
^ ^ ^
(12.5)
An estimate for the mean time to failure (or mean time to repair or mean time to support) can be directly obtained from the sample mean. That is,
MTTF = ∑
^
ti i =1 n
n
(12.6)
Estimate for the variance of the failure distribution can be obtained from the sample variance, that is
(t i − MTTF ) 2 s =∑ n −1 i =1
2 n ^
(12.7)
Estimate for MTTR (MTTS) and Variance of time to repair distribution (time to support distribution) can be obtained by replacing failure times by repair times (support times) in equation (12.6) and (12.7) respectively.
Confidence Interval
It is always of the interest to know the range in which the measures such as MTTF, MTTR and MTTS might lie with certain confidence. The resulting interval is called a confidence interval and the probability that it contains the estimated parameter is called its confidence level or confidence coefficient. For example, if a confidence interval has a confidence coefficient equal to 0.95, we call it a 95% confidence interval. To derive a (1α) 100% confidence interval for a large sample we use the following expression: MTTF ± zα / 2 (
^
σ n
)
(12.8)
215
Where zα/2 is the z value (standard normal statistic) that locates an area of α/2 to its right and can be found from the normal table. σ is the standard deviation of the population from which the population was selected and n is the sample size. The above formula is valid whenever the sample size n is greater than or equal to 30. The 90%, 95% and 99% confidence interval for MTTF with sample size n ≥ 30 are given below:
^ σ 90% confidence MTTF ± 1.645 × n ^ σ 95% confidence MTTF ± 1.96 × n ^ σ 99% confidence MTTF ± 2..58 × n
(12.9)
(12.10)
(12.11)
When the number of data is small (that is when n is less than 30), the confidence interval is based on t distribution. We use the following expression to calculate (1α)100% confidence interval.
^ s MTTF ± tα / 2 n
(12.12)
where tα/2 is based on (n1) degrees of freedom and can be obtained from t distribution table (refer appendix). Example 12.1 Time to failure data for 20 car gearboxes of the model M2000 is listed in Table 12.1. Find: 1. Estimate of failure function and reliability function. 2. Plot failure function and the reliability function. 3. Estimate of MTTF and 95% confidence interval. Table 12.1. Failure data of gearboxes in miles 1022 9007 14363 19030 SOLUTION: The failure function and reliability function can be estimated using equations 12.2 and 12.3. Table 12.2 shows the estimated values of failure function and reliability function. 1617 10505 15456 19365 2513 11490 16736 19596 3265 13086 16936 19822 8445 14162 18012 20079
216
Table 12.2. Estimate for failure and reliability function. Failure data 1022 1617 2513 3265 8445 9007 10505 11490 13086 14162 14363 15456 16736 16936 18012 19030 19365 19596 19822 20079
^
^
F (t i )
0.0343 0.0833 0.1324 0.1814 0.2304 0.2794 0.3284 0.3774 0.4264 0.4754 0.5245 0.5735 0.6225 0.6716 0.7206 0.7696 0.8186 0.8676 0.9167 0.9657
R (t i )
0.9657 0.9167 0.8676 0.8186 0.7696 0.7206 0.6716 0.6225 0.5736 0.5246 0.4755 0.4265 0.3775 0.3284 0.2794 0.2304 0.1814 0.1324 0.0833 0.0343
The failure function and the reliability function graph are shown in Figure 12.1 and 12.2 respectively.
1 Fa ilure function 0.8 0.6 0.4 0.2 0 0 5000 10000 15000 20000 25000
Age of the ca r(in m ile s)
Figure 12.1 Estimate of failure function for the data shown in Table 12.1
217
Example 12.Estimated reliability function 1 0.3.5 ± 2.6 0.025 for (n1) = 19 is given by 2. From ttable the value of t0.98.2 Estimated reliability function for the data given in Table 12.16 miles n −1 i =1 n ^ As the sample data is less than 30. Time to repair data 28 30 31 33 35 53 56 58 59 61 71 72 74 75 79 90 92 94 95 97 218 . The 95% confidence level for MTTF is given by: ^ s = 12725. i =1 20 20 Estimate for the standard deviation is given by (t − MTTF ) 2 s= ∑ i = 14827. we use equation (12.8 0. Table 12.5 miles.3. Find 95% confidence level for MTTR. 19845. the 95% confidence interval for MTTF is (5605.4 0.01). Find the cumulative time to repair distribution and mean time to repair.2 0 0 5000 10000 15000 20000 25000 Age of the car (in miles) Figure 12.093.2 Time taken to complete repair tasks for an item is given in Table 12.16 / 19 ) MTTF ± tα / 2 n That is.093(14827.12) to find the 95% confidence level.2 The estimate for mean time to failure is given by: MTTF = ∑ ^ ti = 12725.
95% confidence level for MTTR is given by ^ 23. 3 i − 0.Maintainability function for the data given in Table 12. 3 = n + 0.3 shows the estimated maintainability function.43.7 hours i =1 40 40 Standard deviation for repair time is given by (t − MTTR) 2 s = ∑ i = 23.43 hours n −1 i =1 2 n ^ Since n > 30.2 0 0 20 40 60 Tim e 80 100 120 Figure 12.96) MTTR ± 1. 4 40. we use equation (12.4 0.3.4 Figure 12.40 41 44 49 51 65 67 68 69 70 81 82 84 85 89 99 100 103 108 110 Maintainability function can be estimated using following expression: M (t i ) = ^ i − 0.3.6 0.96) = 69.7 ± (1. 76.10) to calculate 95% confidence interval for MTTR. Mean Time to Repair is given by: MTTR = ∑ ^ ti = 69.43 s = (62.96 40 n 219 . M a inta ina bility Function 1 0.8 0.
i ^ f (t ) = F ( X max.Analysis of Grouped Data Often failure data is placed into time intervals when the sample size is large.i ) = ^ ^ k =i +1 ∑ ni + 0.i is the lower bound of the ith interval and Xmax. Let n1.i ) X max. i = x min + (i − 1) × LI X max.i +1 ) − F ( X max. The failure data are classified into several intervals.i +1 − X max.4) × ( X max.14) where xmax is the maximum recorded failure time and xmin is the minimum recorded failure time.3 × log 10 (n) (12. n2 . The length of each interval.15) Estimate for the reliability function R(t) is given by: R( X max. The number of intervals.i ) = 1 − F ( X max.i+1 < t < Xmax.16) Estimate for the failure density is given by: For Xmax.i ) = ^ k =1 ∑ nk − 0. LI. … nn be the number of items that fail in the interval i.i ^ = ni +1 ( n + 0.4 n (12. depends on the total number of data n. NI.3 n + 0.4 i (12.i = x min + i × LI Xmin.13) NI denotes that the value is rounded down to the nearest integer.i is the upper bound value of the ith interval. Then the estimate for cumulative failure distribution is given by F ( X max.i ) The MTTF is estimated using the expression: 220 . is calculated using: LI = (x max − xmin ) NI (12. Following equation can be used as guidance for determining the suitable number of intervals: NI = 1 + 3.i +1 − X max.7 n + 0. The lower and upper bound of each interval is calculated as follows: X min.
88 1272.13).5 shows the various calculations associated in computing the mean time to support.50.34 34. Analysis of grouped data given in example 12. Calculate the Mean Time to Support (MTTS).18.17 34.66.76 90.MTTF = ∑ ^ NI X med .92 Xmed.99 ni 6 10 11 5 9 14 xmed.4.55 466.51 50.68 .3 × log 10 (55) = 6.17) where Xmed. Time to support data 3 26 99 32 31 56 76 2 44 78 9 79 3 46 34 24 89 37 5 67 56 45 39 46 86 66 45 39 46 86 67 78 77 99 75 87 88 93 47 33 89 89 21 77 55 99 90 24 79 22 4 92 29 89 44 SOLUTION: First we need to find the number of groups using equation (12.i 10.18) Example 12.67 292.i − MTTF ) 2 × i =1 NI ^ ni n (12.25 42.i × ni 60.i × ni n i =1 (12.08 26.I .95 MTTS is given by: 221 .5.97 672.85 82.17 = 6 NI Table 12.i) 2 . Estimate for sample variance is given by s 2 = ∑ ( X med . The number of intervals is given by: NI = 1 + 3.42 58.82.17 18. Table 12.3 Results of 55 observed values of the duration of support tasks in hours are given in Table 12.4.59 74.68 66.51 262.xmax.34 .3 i 1 2 3 4 5 6 LI (xmin.51 .i is the midpoint in the ith interval and nk is the number of observed failures in that interval. Table 12.74 = 6 The length (range) if each interval (group) is given by: LI = xmax − x min 99 − 2 = 16.85 .
19) where. R0 is zero and Si is the sequence number of the ith failure.50. 3404*. on a fixed number of failures or at some arbitrary point in time. 2665*. at the time of suspension (censor) are known. 2173. 2. As a quick check. If there is no suspensions. 3716 222 . provided the times at the time of failure or. the reason for terminating the test is not important. We denote ti to represent a complete data and ti* to denote a censored time.3 n + 0. Ri1 is the adjusted rank of the (i1)th failure. that is the previous failure. The only difference between the estimation of parameters in complete data and the censored data is the calculation of median ranks. 3008. Estimate the reliability and failure function (* indicates that the data is a censored data) 2041.i × ni n i =1 =∑ 6 X med . only that the failure occurred sometime between the last inspection and the current age later. the adjusted rank of the ith failure will always be less than or equal to the sequence number and at least 1 greater than the previous adjusted rank.4 Example 12. We will look at cases in which we do not know the exact time. 3490*. reliability function when the data is multiple censored. Now we will need to adjust the ranks in order to take account of the components that have not failed.i × ni 55 i =1 = 55. For each failure calculate the adjusted rank as follows: Ri = Ri −1 + n + 1 − Ri −1 n + 2 − Si (12. In practice. Now we discard the suspended times as it is only the (adjusted rank) of the failures with which we are concerned. The mechanism for censoring may be based on a fixed age. the adjusted rank will be equal to the sequence number as before. 3424. 2248*. ANALYSIS OF CENSORED DATA In many cases. 2567*. We will assume that the times of failure are known precisely. Ri is the adjusted rank of the ith failure. The rank adjustment is done in the following two steps: 1.4 The following data were observed during the data capturing exercise on 12 compressors that are being used by different operators. In this section we derive estimates for failure function. the complete data may not be available due to the reasons such as all the items may not have failed or the manufacturer may wish to get interim estimates of the reliability etc.MTTS = ∑ ^ NI X med . Sort all the times (failures and suspensions) in ascending order and allocate a sequence number i starting with 1 for the first (lowest) time and ending with n (the sample size for the highest recorded time).06 12. These adjusted ranks are then substituted into the Benard's approximation formula to give the median rank and the estimate for cumulative probability is given by: F (t i ) = ^ Ri − 0. 2271. 3091.
SOLUTION: We need to calculate the adjusted rank of the failure times using equation (12.19).8097 R(t i ) 0. The estimated failure and reliability functions are shown in Table 12.5468 0.34 0. then the failure and reliability function can be estimated using equations (12.0565 2 2173 2 2 0.6605 0. once this is done. be borne in mind that failures do not occur in accordance with a given distribution. Table 12.6 Estimated failure and reliability function Si ti j Rj = Rj1 + [(n+1– Rj1) F (t i ) / (n+2 – Si) ] 1 2041 1 1 0. Table 12. Failure data for 50 tyres 1022 1617 2513 3265 8445 9007 10505 11490 13086 14162 14363 15456 16736 16936 18012 19030 19365 19596 19822 20079 20208 20516 20978 21497 24199 24582 25512 25743 26102 26163 26530 28060 28240 28757 28852 29092 29236 29333 30620 30924 31507 33326 33457 35356 35747 36250 36359 36743 36959 38958 223 .4532 * 9 3404 10 3424 6 7.5960 11 3490* 12 3716 7 10.51.9435 0. maintenance and supportability characteristics is using a theoretical probability distribution.69 0. In this section we will look at a graphical method that can be used to not only to fit distributions to given data but also help us determine how good the fit is. FITTING PROBABILITY DISTRIBUTIONS GRAPHICALLY The traditional approach for measuring reliability.8630 0.3395 8 3091 5 5.1903 12.7742 0.1370 * 3 2248 4 2271 3 3. To illustrate the graphical approach we use the following failure data observed on 50 tyres. It should however. These are merely convenient tools that can allow us to make inferences and comparisons in not just an easier way but also with known levels of confidence.1 0.51 0.4040 0.3) respectively.2) and (12.2258 * 5 2567 6 2665* 7 3008 4 4.92 0.7.6.
1 0 0 5000 10000 15000 20000 TimetoFailure 25000 30000 35000 40000 Figure 12. Now. Sorting the timestofailure in ascending order will give us the 'x' values so all we need is to associate a cumulative probability to each value. What it does not tell us. This indicates that the exponential distribution is not a very good fit whereas the normal is certainly better. is how much better or. two standard approaches to fit the data to a probability distribution graphically: to use “probability paper” or to transform either the “x” or “y” (or both) data so that the resulting graph would be a straight line if the data were from the given distribution. F(ti)]. in fact. This is done using the median rank approach discussed earlier.2 0. 1 0. however.6 Probability Tyre Data Exponential Normal 0. this only applies to straight line fits.7 0. Actually both methods are essentially the same because to create probability paper the axes have been so constructed as to produce straight lines plot if the data is from the given distribution. Similarly we could use the KolmogorovSmirnov test but this really only tells us whether the there is a significant difference between the data and that which would be expected if the data were exponentially or normally distributed. whether another distribution gives an even better fit. indeed. we can plot the values [ti. In Figure 12.5 0.8 0. If we can determine the necessary transforms then we can easily construct the probability paper. A measure of how good the curve fits the data would be the correlation coefficient but.4 Tyre Data compared to Exponential and Normal Distributions The two additional lines on this graph have been plotted to show what an exponential distribution (with the same mean as the sample) would look like and similarly for a normal distribution with the sample mean and standard deviation.2).4 0. There are.4 we can see the result of this for the 50 tyre timetofailure. that is 'y' axis values are given by the cumulative failure probabilities calculated using the equation (12.To draw a graph we obviously need a set 'x' and 'y' coordinates.3 0. Fitting an exponential distribution to data graphically The cumulative probability density function for the exponential distribution is given by 224 .9 0.
5 is an example of “Exponential Graph Paper” (for the failure date from Table 12. If we replace F(t) with p then we get p = 1 − exp( −λt ) Rearranging and taking natural logarithm we get ln[ 1 ] = λt 1− p (12.7). we can. we find the time to failure from the paper for which the percentage failures is 63. omit the expression for t < 0.632. That is. Figure 12.5 Data Plotted on Exponential Graph Paper If the data forms a straight line in the exponential probability paper. t ≥0 Since we are only considering positive failure times. The xscale is linear. t < 0 F (t ) = 1 − exp( −λt ). then we can find the value of MTTF by using the relation F(MTTF) = 0.0. without loss of generality. 225 .2. The yscale is given as percentages rather than probabilities.20) This is a linear function in t such that the slope of the line is the reciprocal of the MTTF. 99 Cumulative Probability (%) 95 90 80 70 60 50 40 20 0 5000 10000 15000 20000 Times to Failure 25000 30000 35000 40000 Figure 12.
6 shows how the tyre example failure times (and their respective median ranks) would appear on “normal paper”. Figures 12.7 and 12.5 and F(µ+σ) = 0. 226 .84. Now we can plot this value(as the y coordinate) against the corresponding timetofailure (as the x coordinate). Figure 12. Such paper is becoming increasingly more difficult to obtain commercially. σ for any given value of p (F(t)) either from tables or.Fitting a Normal Distribution Graphically We will now see how good a fit the normal distribution gives. however. F(µ) = 0. be created using a proprietary spreadsheet package.7. Fitting a LogNormal Distribution Graphically Essentially the lognormal distribution is the same as a normal distribution excepting that the (natural) logarithm of the xvalues are used in place of the actual values.8 show lognormal plot for the data given in Table 12. using the NORMSINV function in MicroSoft™ Excel®. The value of µ and σ can be found by using the relation. for example. Again we can plot the timestofailure on special normal (probability) paper.6 TimestoFailure plotted on Normal Paper The cumulative density function for the normal distribution is not as simple to transform to a linear form as the exponential. 99 95 90 80 70 60 50 40 30 20 10 5 1 0 10000 Times to Failure 20000 30000 40000 Figure 12. It can. we can obtain the standardised normal variable z = ( x −µ ). t F (t ) = p = ∫ 1 2π σ x−µ − 1 2 σ e 2 dx −∞ However.
99 95 90 80 70 60 50 40 30 20 10 5 1 1000 10000 100000 Figure 12.8 Fitting a LogNormal Distribution Graphically 227 .5 8 8.0614x .444 R2 = 0.10.5 9 9.5 7 7.5 10 10.5 11 1 2 3 4 Ln(Time to Failure) Figure 12.7346 Normal Ordinate 0 6 6.7 TimestoFailure plotted on LogNormal Paper 3 2 1 y = 1.
776 which is considerably lower than the mean from the previous graphs but. this is because it is the geometric mean (the nth root of the product of the TTFs) and not the arithmetic mean with which we are more familiar. Fitting a Weibull Distribution Graphically The cumulative density function of the Weibull distribution is similar to that of the exponential. To get an estimate of the scale parameter (η) we need to carryout a transform on the intercept: η=e − cβ where c is the intercept of the regression line with the xaxis. Figures 12.e for t ≥ 0 By rearranging and taking natural logarithms t − ln(1 − p) = η β which is still not in a linear form so we have to take logs again to give: ln( − ln(1 − p)) = β ln(t ) − β ln(η) So if we plot ln(ln(1p)) against ln(t) an estimate of the shape parameter (β) of the Weibull will be given by the slope of the straight line drawn through the plotted points. 228 .7. indeed the latter is the (mathematically) degenerative form of the former.9 and 12. The mean in this case is 18. for t < 0 0 β F (t ) = p =  t η 1.Here the plotted points form a concave curve to which the straight line is not a particularly good fit although it is still better than the exponential fit.10 shows Weibull plot for the data given in Table 12.
935 R2 = 0.5 7 7. 229 .5 10 10. The slope (1.5 11 1 y = 1.14.4608x .5 8 8.9 TimestoFailure Fitted on Weibull Paper 2 1 0 6 Transformed Probability 6.48) indicates that there could be a certain amount of agerelatedness to the failures.99 95 90 80 70 6 50 40 30 20 10 5 1 1000 100000 10000 Figure 12.10 Fitting a Weibull distribution graphically Again the Weibull distribution does not give as good a fit as the normal (distribution) but it is better than either the exponential or the lognormal.5 9 9.883 2 3 4 5 Log(Time to Failure) Figure 12.
If we construct a line such that it passes through the point x. The measure of how well this line fits the data is given by the correlation coefficient. linear regression finds the bestfit straight line for the set of points (x. The same holds for the horizontal distances provided that the line is not parallel to the xaxis.12.11 illustrates the least square regression. y where x is the mean of the x values and y is the mean of the y values then the sum of the distances between each point and the point on the line vertically above (ve) or below (+ve) will always be zero (provided the line is not parallel to the yaxis). Y points that have been measured.52. To check whether there is a linear relationship between the dependent variable and the independent variable.21) 230 . Figure 12. The objectives of linear regression are: 1. y). To estimate the constants and of the best fit y = a + bx. y Slope = b a x Figure 12.11 Least square regression. That is. 3. To find the best fit straight line for a given set of data points. In reliability analysis. The simplest regression model is the one that relates the variable y to a single independent variable x (linear regression model). This means that any line passing through the means (in the way described) will be an unbiased estimator of the true line. Linear regression provides predicted values for the dependent variables (y) as a linear function of independent variable (x). that the x’s are known exactly and that the “errors” in the y values are normally distributed with mean 0 then it can be shown that the values of a and b which minimises the expression: 2 ∑ ( y i − a − bxi ) n ( ) i =1 (12. The standard method for linear regression analysis (fitting a straight line to a single independent variable) is using the method of least squares. Least square regression is a procedure for estimating the coefficients ‘a’ and ‘b’ from a set of X. REGRESSION The models used to relate a dependent variable y to the independent variables x are called regression models. If we now assume that there is a linear relationship between the x’s (x ∈ X) and y’s (y ∈ Y). the set X is the set of time to failures (or function of TTF) and set Y is their corresponding cumulative probability values (or function of cumulative distribution). 2.
25) i =1 2 ∑ xi n Note this line does not pass through the means (unless it is a perfect fit).23) 2 a=∑ n x yi − b∑ i i =1 n i =1 n n (12.e. y = a + bx is the equation of the line giving: n ∑ xi y i − ∑ xi ∑ yi i =1 n n n n b= n ∑ xi2 i =1 − ( ∑ xi ) i =1 i =1 n i =1 (12. By replacing each x with a y and each y with an x we can perform “x on y” regression (which assumes the errors are in the xvalues). 231 .24) Note also that these expressions are not symmetrical in x and y. The formula quoted here gives what is called “y on x” regression and it assumes the errors are in the yvalues.e. If c is the estimate of the intercept so obtained and d is the estimate of the slope then to get estimates of a and b (the intercept and slope of the original graph): b= c 1 and a = − d d Note: unless the points are collinear. the intercept is zero). The expression (yi – a – bxi) gives the vertical distance between the point and the line. the “x on y” estimates will not be the same as the “y on x” estimates. one can show that the values of a and b can be found by solving the following equations: na + b ∑ xi = ∑ y i i =1 i =1 n n n n n (12. Cutting out lot of algebra. In the special case where you want to force the line through the origin (i.22) ‘a’ is the estimate of the intercept (of the line with the yaxis) and ‘b’ is the estimate of the slope – i.21) a ∑ xi + b ∑ xi2 = ∑ xi yi i =1 i =1 i =1 (12.Will give the best fit. the least squares formula for the slope becomes: b= i =1 n ∑ xi y i (12.
27) is a linear function. normal and lognormal. ln[ provides a straight line. The correlation coefficient. The best fit distribution is the one with maximum r value (close to one). The cumulative distribution of exponential distribution is given by: F (t ) = 1 − exp( −λt ) that is.Correlation Coefficient A measure of the dependence between two variables is given by the correlation coefficient. A value of +1 or –1 means that x and y are exactly linearly related.26) 2 The correlation coefficient always lies between –1 and +1. but r = 0 does not mean that x and y are independent. r is given by: n ∑ xi y i − ∑ xi ∑ y i n ∑ xi2 i =1 n n n n r= − ( ∑ xi ) × n ∑ 2 i =1 i =1 n i =1 i =1 n i =1 y i2 − ( ∑ yi ) i =1 n (12. for an exponential distribution. In the former case y increases as x increases but for r = 1. the plot of ( t . Thus.23) we get: 232 . Weibull. F(ti)) such a way that. The one with highest correlation coefficient is selected as the best. Here ti is the observed failure times and F(ti) is the estimated cumulative distribution function. yi) in equation (12. y) and the corresponding parameters for different distributions are listed given in the following sections. Note that if x and y are independent then r = 0. To find the best fit. it gives a straight line.27) 1 ]) 1 − F (t ) . regression analysis is carried out on the popular distribution such as exponential.28) (12. we set: xi = t i y i = ln[ 1 ] 1 − F (t i ) (12. tn are the observed failure times. The coordinates (x.29) Substituting (xi . then to fit this data into an Equation (12. if t1. t2. we transform the coordinates (ti. Thus. ln[ 1 ] = λt 1 − F (t ) (12. Linear Regression for Exponential Distribution To fit a data to an exponential distribution. exponential distribution. y decreases as x increases. when plotted.
34. various calculations are tabulated in Table 1 − F (ti ) The value of b is given by: b= i =1 n ∑ xi y i 2 ∑ xi i =1 n = i =1 ∑ ti × ln[ n n 1 ] 1 − F (t i ) = 0.1196 1.9513 yi = ln[1 / (1F(ti))] 0.8.5 The following failure data were observed on Actuators.6736 0.0486 0. 230 SOLUTION: First we carry out least square regression on t i .1875 0. 14. 61.01126 = 88.1180 0.7651 0.9267 1.73.9666.8819 0. 233 .1366 3.8125 0.3958 0.2076 0. ln[ 12. Fit the data to an exponential distribution and find the MTTF and the correlation coefficient. 66.6739 2.7430 0.b= i =1 n ∑ xi y i (12.4652 0. Table 12. The corresponding correlation coefficient is 0.0498 0. 27.2969 0.5 i 1 2 3 4 5 6 7 8 9 10 11 12 13 14 ti (= xi) 14 27 32 34 54 57 61 66 67 102 134 152 209 230 F(ti) 0.5039 0. 102.5347 0.8. 152.3951 0.6260 0.3588 1. Example 12. 32. for exponential distribution b = 1/MTTF. 67.01126 2 ∑ ti i =1 MTTF is given by 1/b = 1/0. Regression analysis for the data in example 12.30) i =1 2 ∑ xi n Note that. 209.2569 0. 57.0239 1 ] . 54.6041 0.3263 0. 134.1256 0.
3665 .0445 3.3245 0.8508 0. Thus to fit the data to a Weibull 1 − F (t ) distribution.8233 .8828 4. 39.5 0. 129. 132.0.6754 0.6 i 1 2 3 4 5 6 7 8 9 10 11 ti 17 21 33 37 39 42 56 98 129 132 140 F(ti) 0.1. which are presented in Table 12.0.0.8598 4.6434 1.9385 xi = ln(ti) 2.0614 0.3648 0. it is evident that the shape and scale parameters of the distribution are given by: β =b η = exp( − a / β ) Example 12. which is a linear function.32).6635 3.7581 . 33. 56.3082 .9.4122 0.1.31) and (12.34) (12.7631 0.Linear Regression for Weibull Distribution Cumulative distribution of Weibull distribution is given by: t F (t ) = 1 − exp( −( ) β ) η That is. 140 (12.9416 Yi = lnln(1/1F(ti)) .5877 0.2368 0.6320 . 98. ln[ln( 1 )] = β ln(t ) − β ln(η ) .4965 3. 37.0. Table 12.7376 4. we construct the least square regression. Weibull regression for the data in example 12.6 Construct a least square regression for the following failure data: 17.9354 .31) (12.5849 4.0261 234 .8332 3.0253 4. 42.32) From least square regression.2.1491 0.6109 3. 21.1209 0. we set: xi = ln(t i ) y i = ln[ln( 1 )] 1 − F (t i ) (12.1180 0.9.35) SOLUTION: Making use of equations (12.
802853.432788.35).515517. Linear regression for Normal Distribution For normal distribution. η = 76.010328. Now for regression. C2 = 0. One can also use the following expression that gives polynomial approximation for zi.34) and (12. we get β = 1. d3 = 0. we set xi = ti and yi = zi = Φ1 [F(ti )].Using equations (12. d1 = 1.4355.36) Which is a linear function.001308 The estimate for µ and σ are given by µ =− a 1 and σ = b b Example 12.37) yi = P − C 0 + C1 P + C 2 P 2 1 + d1 P + d 2 P 2 + d 3 P 3 (12. F (t ) = Φ ( t−µ ) = Φ( z ) σ Now z can be written as: z i = Φ −1[ F (t )] = ti − µ ti µ = − σ σ σ (12. C1 = 0.7 Fit the following data into a normal distribution 235 . xi = t i P = ln[ 1 [1 − F (t i )] 2 ] (12.54 and the correlation coefficient r = 0.9133. d2 = 0.38) where C0 = 2.189269. The value of z can be obtained from standard normal distribution table.
75. Linear Regression for Lognormal Distribution For lognormal distribution we set: xi = ln(t i ) P = ln[ 1 [1 − F (t i )] 2 ] (12. σ = 54. d1 = 1.7 i 1 2 3 4 5 6 7 ti 62 75 93 112 137 170 185 F(ti) 0.001308 236 .40) where C0 = 2.0.189269.7394 1.10. 185 SOLUTION: Table 12.62. 137.71.7702 0.6351 0.7302 . C1 = 0.1.2297 0.0945 0.515517.3132 The estimate for µ = 118.3434 0 0.3450 0. 93.10 gives various computations involved in regression.9701.010328.9054 zi = P – (c0 + c1 P + c2 P2 / 1 + d1P + d2P2 + d3P3) . d2 = 0. C2 = 0.5 0. Normal regression for example 12. 112.0.05 and the correlation coefficient r = 0. 170. Table 12.432788. d3 = 0.39) yi = P − C 0 + C1 P + C 2 P 2 1 + d1 P + d 2 P 2 + d 3 P 3 (12.2693 .802853.3648 0.
U Dinesh Kumar Indian Institute of Management Bangalore 237 .Total Cost of Ownership (TCO) and Life Cycle Cost Models Prof.
3 Life of the Asset 1.2 Initial Capital Cost 1.7 Uncertainties and Sensitivity Analysis 1.14 LCC Models in Designing for Logistic Support 2.10 PRICE Life Cycle Costing System 2.11 Equipment Designer s Cost Analysis System (EDCAS) 2.1 Life Cycle Cost/Total Cost of Ownership Concept 2.3 Life Cycle Cost Technique 2.2 Definitions of Life Cycle Cost/Cost of Ownership 2.5 Taylor s LCC Model 2.7 Roskam s LCC Model 2.1 Introduction.8 Fabrycky and Blanchard s LCC Model 2.6 Disposal Cost 1. 1. Survey of Existing Literature 2.5 Operation and Maintenance Cost 1.4 The Discount Rate 1.CONTENTS 1.8 Summary 2.13 LCC Models for Labour Factor 2.12 LCC Models Using Markov Chain 2. Life Cycle Cost and Total Cost of Ownership 1.6 Raymer s LCC Model 2.4 Review of Life Cycle Cost Models 2.15 Applications of LCC/Cost of Ownership Models 34 3 3 5 5 6 6 6 6 7 09 09 12 13 16 23 25 25 26 28 28 28 30 31 31 References 35 238 .9 Burn s Life Cycle Cost Model 2.
239 .
that is. by funding the project and operating the related product. In the military sector the consumer. LCC analysis is applied routinely to military projects. the use of life cycle cost and cost of ownership has extended to other areas of the public and private sectors.1. the main focus is on TCO where related costs. budget planning. and range of other activities that occur over the life of complex technological equipment. The objectives of LCC/TCO are (Flanagan and Norman. covering acquisition (purchase or lease).1 Introduction “Value for Money” has become one of the important criteria in an increasingly competitive business environment. The term LCC analysis is rarely used in the commercial sector. From its origins in defence equipment procurement in the US in early 1960s. Ø To assist in the effective management of completed projects. operation. maintenance and support are borne by the customer. Life Cycle Cost and Total Cost of Ownership ___________________________________________ 1. Instead. In addition. 1983): Ø To enable investment options to be more effectively evaluated. cost control. Ø To consider the impact of all costs rather than only the initial capital costs. LCC and TCO are being used to assist in decisionmaking. the customer can also incur costs when the product is not available for use. ‘down time costs’. Life Cycle Cost (LCC) and the Total Cost of Ownership (TCO) are two important financial measures that are used for decision making in acquisitions. It is important to consider the difference between LCC and TCO. 240 . essentially bears the total life cycle cost covering the major cost elements in all stages of a product’s life cycle.
In order to achieve these objectives the following elements of cost of ownership have been identified (Woorward. military or space projects (Knotts. 6. The US Department of Defence has formally used the concept of life cycle cost in weapon system acquisition since the early 1960s through life cycle costing and life cycle cost analysis. 2. Testing and Certification. 5. 1998). Design and Development. The number of phases depend on the nature of the project. Commonly used phases are: 1.Ø To facilitate choice between competing alternatives. 1997): Ø Initial capital costs Ø Life of the asset Ø The discount rate 241 . purpose and whether they are applied to commercial. Requirements (Functional Specification). 85% by the end of system definition and 95% by the end of full scale development. Disposal It is reported by the US Department of Defence that 70% of weapon system life cycle cost is committed by the end of concept studies. Maintenance and Support. 7. Production. In the Defence industry the system’s life cycle is divided into various phases. 4. which allow proper planning and control of a project. The cost of ownership approach identifies all future costs and reduces them to their present value by use of the discounting techniques through which the economic worth of a product or product options can be assessed. 3. Operation. Concept/Feasibility Studies.
1. 1. fees. Basically. Physical life – the period over which the asset may be expected to last physically. Economic life – the period until economic obsolescence dictates replacement with a lower cost alternative. (2) acquisition/finance costs. Ferry et al (1991) has defined the following five possible determinants of an asset’s life expectancy: Functional life – the period over which the need for the asset is anticipated. to when replacement or major rehabilitation is physically required.Ø Operating and maintenance costs Ø Disposal cost Ø Uncertainty and sensitivity anal