Академический Документы
Профессиональный Документы
Культура Документы
VOLUME I
Handbook
Books by Dimitri B. Kececioglu, Ph.D., P.E.
Thii O ni
9Q6E-A6X-67DQ
HOW TO ORDER THIS BOOK
by phone: 7 1 7 290- 1660 or 7 17-394-4583, 8am-5pm Eastern Time
BY FAX: 717-509 6100
by mail Order Department
DEStech Publications, Inc.
1148 Elizabeth Avenue #2
Lancaster, PA 17601, U.S.A.
by credit card: VISA, MasterCard
by www site: http://www.destechpub.eom
Maintainability,
Availability, &
Operational
Readiness
Engineering
Handbook
VOLUME 1
0
DESteclj Publications
Maintainability, Availability & Operational Readiness Engineering
Handbook, Volume 1
DEStech Publications, Inc
1 148 Elizabeth Avenue #2
Lancaster, Pennsylvania 17601 U.S.A.
ISBN No 1-932078-05-3
To my wonderful wife Lorene,
daughter Zoe,
and son John.
VOLUME 1
TABLE OF CONTENTS
PREFACE xxix
CHAPTER 1- MAINTAINABILITY ENGINEERING 1
1.1- HANDBOOK OBJECTIVES 1
1.2- RELIABILITY AND MAINTAINABILITY
ENGINEERING OBJECTIVES AND
BENEFITS 4
2.1.1- WSEIAC 21
2.1.2- ARINC 22
2.2- A COMPREHENSIVE DEFINITION
OF RELIABILITY 26
2.3- A QUANTIFICATION OF SYSTEM
EFFECTIVENESS 27
EXAMPLE 2-1 29
REFERENCES 36
CHAPTER 3- MAINTENANCE 37
3.1- MAINTENANCE DEFINED 37
EXAMPLE 5-2 77
SOLUTION TO EXAMPLE 5-2 77
EXAMPLE 5-3 80
SOLUTION TO EXAMPLE 5-2 80
PROBLEMS 84
CHAPTER 6- MAINTAINABILITY AND ITS QUANTIFICATION ... .91
6.1- MAINTAINABILITY DEFINED 91
6.2- THE EXPONENTIAL CASE 93
EXAMPLE 6-1 96
SOLUTION TO EXAMPLE 6-1 97
6.3- THE REPAIR RATE, /i 97
6.4- THE MEAN TIME TO REPAIR, MTTR 97
EXAMPLE 6-2 99
SOLUTION TO EXAMPLE 6-2 99
6.5- MEAN ACTIVE CORRECTIVE
MAINTENANCE TIME, UTc 99
6.6- MEAN ACTIVE PREVENTIVE
MAINTENANCE TIME, t^ 100
6.7- MEAN ACTIVE CORRECTIVE AND
PREVENTIVE MAINTENANCE TIME, t]j 100
6.8- EQUIPMENT REPAIR TIME, ERT 101
6.9- GEOMETRIC MEAN TIME
TO REPAIR, MTTRG 101
6.10- MAXIMUM MAINTENANCE TIME, tM. 102
EXAMPLE 6-3 102
SOLUTIONS TO EXAMPLE 6-3 103
6.11- THE LOGNORMAL CASE 106
EXAMPLE 6-4 107
x CONTENTS
PROBLEMS 139
CHAPTER 8- MAINTAINABILITY ENGINEERING
SPECIFICATIONS AND
THEIR APPLICATIONS 147
PROBLEMS 204
REFERENCES 334
CHAPTER 10- PERIODIC REPLACEMENT POLICIES 335
10.1- ORDINARY PERIODIC REPLACEMENT
POLICY-POLICY III 335
EXAMPLE 10-1 336
PROBLEMS 354
REFERENCES 357
REFERENCES 455
PROBLEMS 525
REFERENCES 531
REFERENCES 696
PROBLEMS 724
REFERENCES 738
CHAPTER 17- RELIABILITY OF COMPONENTS
WITH A POLICY OF REPLACING
THOSE THAT FAIL BY A
PRESCRIBED OPERATING TIME 739
including the use of the Markov chains approach with constant and
nonconstant failure and repair rates.
The combination of reliability, which may be quantified if the mean
time between failures (MTBF) is known, and of maintainability, which
may be quantified if the mean time to restore the equipment to success
ful function (MTTR) is known, yields the steady state availability of
any equipment. The availability of equipment is a very important mea
sure, because it determines the percent of their uptime; consequently,
the percent of their operational time the equipment is available to give
the desired level of output or production. The overhead for equipment
being essentially fixed, the more available the equipment is to deliver
the required output or to manufacture a product, the lower will be the
overhead cost per unit of output or per unit of manufactured product.
The intrinsic, inherent, or instantaneous, as well as a great variety of
steady state availabilities, are quantified in these handbooks and their
applications are illustrated by numerous worked-out examples.
In addition to assuming that the equipment is available to provide
the desired function after the start of its operation, it has to be ready
to start to operate at a desired point in time, or when the call for it
to start to operate arrives. The concept that quantifies the probability
that the equipment will be ready to start its function, when called upon
to do so, needs to be developed and quantified. This is accomplished by
the concept of operational readiness, which is covered extensively
in these handbooks.
An all-encompassing concept which quantitatively combines op
erational readiness, mission reliability and design adequacy is
system effectiveness. This concept needs to be developed to assure
that not only the equipment starts its operation when needed, it also
completes its mission, or function, satisfactorily and performs all of its
designed-to functions as specified. Methods for quantifying system
effectiveness are developed and illustrated in these handbooks for
quick implementation thereof.
After the equipment is designed and built, its designed-in maintain
ability needs to be demonstrated. MIL-STD-471 gives many methods
of achieving this. These methods are covered in detail, and all demon
stration models are derived and illustrated by many examples in these
handbooks.
To assure that the equipment will meet their maintainability goals,
methodologies need to be developed to predict their maintainabil
PREFACE xxxi
PA 17601 - 4359, Vol. 1, 720 pp. and Vol. 2, 568 pp., and "Reliability
& Life Testing Handbook" by Dr. Kececioglu, published by DEStech
Publications, Inc., 1148 Elizabeth Ave. #2, Lancaster, PA 17601 -
4359, Vol. 1, 950 pp., 2002 and Vol. 2, 900 pp., 2002.
Each chapter has numerous practical examples, completely worked
out and necessary computer programs given. Problems to be worked
out by students and practitioners are given at the end of each chapter,
as well as complete references. It is recommended that those who teach
reliability engineering courses start by teaching out of the two-volume
"Reliability Engineering Handbook," follow it by teaching out of the
two-volume "Reliability & Life Testing Handbook," and then follow it
by teaching out of this, the two-volume "Maintainability, Availability
and Operational Readiness Handbook."
Chapter 1 of Volume 1, of this two-volume handbook establishes
the objectives of this handbook, the overall benefits of an integrated
reliability and maintainability engineering program implemented in in
dustry and government, and covers 22 case histories documenting the
actual benefits derived from the implementation of reliability and main
tainability engineering.
Chapter 2 defines and quantifies system effectiveness, and gives
the relationship between reliability and system effectiveness. It also
defines reliability and maintainability comprehensively and provides
46 practical benefits of implementing reliability and maintainability
engineering.
Chapter 3 defines maintenance, preventive maintenance, cor
rective maintenance; identifies and defines all types of corrective and
preventive maintenance downtimes and their relationship to all other
times associated with the life of equipment. It also covers maintenance
personnel factors and costs; maintenance personnel safety factors; and
maintenance support facilities and equipment.
Chapter 4 discusses 15 maintainability design criteria which mini
mize equipment downtime, increase accessibility to critical, high failure
rate parts in equipment, provide better packaging; identify the correct
placement of labels which contain operating and maintenance instruc
tions so that they can be seen easily, and the correct choice of fastener
design; provide for the correct identification of equipment check points,
numbering of parts, marking of connectors so that they are connected
to the correct receptacle, etc.
Chapter 5 covers downtime distributions, and maintainability en
PREFACE xxxiii
MAINTAINABILITY,
AVAILABILITY AND
OPERATIONAL READINESS
ENGINEERING
1
MAINTAINABILITY ENGINEERING
14. The Air Force placed an order with Hoffman Electronic Cor
poration of Los Angeles, California, to undertake a major pro
duction contract for AN/ARN 21-C TACAN equipment, which
is the airborne portion of a ground-linked, short-range, naviga
tional aid system [7] with a strict requirement of implementing
a complete reliability and maintainability engineering program.
The economic advantages of these "build-it-right-the-first-time"
procurement actions taken by the Air Force and Hoffmann can
be summarized as follows and in Table 1.2 :
14.1 Price savings of $445 per set times 10,000 sets on order or
$4,450,000.
14.2 Service life estimates of 3,000 hr.
12 MAINTAINABILITY ENGINEERING
14.3 8.5 times more reliable; i.e., 17.5 hr versus 150 hr MTBF.
14.4 Maintenance costs per failure of $140.
14.5 In one year on 10,000 sets of navigational systems $70,650,000
was saved.
1 ,906,0 0 18,727,0 0
Total
$
cost, 600
1,800
1,500
1,200
900
aiTABLE
cCost
for
1.3
dtwo
oermnbpt—oaircinaselon maintenance
Annual
ipanels
(based
units).
n2,000
dioncator
5,452,000 14,285,0 0
MTBF
hr.
941
=
$
cost,
MTBF
h33
1Unit
A
=rr
MTBF,
hr 941 331 UnitB $of
6,821,0 0
Unit
cost,
3,227 2,221 sup liers.
$
Supplier
A B 25t
CO
14 MAINTAINABILITY ENGINEERING
16. Table 1.4 illustrates the benefits of trade-off studies which can
not be conducted without reliability and maintainability inputs.
The problem resolved in Table 1.4 is whether to throw away or
repair a failed module from the overall cost point of view. Un
der "Total cost of spares," 60 repairable modules per ten years
are estimated to be needed, yielding $20 X 60 = $1,200, and 50
throwaway modules yielding $150 X 50 = $7,500. This informa
tion is derived from reliability engineering studies. Under "Cost
of repair - active repair time," the MTTR of two hours is needed
yielding 60 repairs/10 years X 2 hr/repair X $15/hr = $1,800.
This MTTR can only be obtained from Maintainability Engi
neering studies. Finally, it is found that the "repair case" will
cost $14,796 versus $8,680 for the "discard case"; consequently,
the "discard case" wins out at great savings.
Module
repair
throw
(single
ap vaway
leircsautsion)
Discard
case $150 $7,500 N/A N/A N/A N/A N/A $1,000 N/A $30 N/A N/A $8,680
Apcostlicable
minputs.
aintainability
Repair
case $200 $1,200 $396 $1,800 $100 $300 none $140 none $60 $600 $10,000 $14,796
shipping
and
packaging
of
Cost tdiagrams,
retc.
ouble-sho ting
COST
ESTIMATED
TOTAL
of isioning, handling,
pCost
rovspares caetc.
taloguing,
eTest
quipment/spares echnical manuals,
ttraining
of
Cost tCost
of
echnical
spare modules)
10%
(Assume repairable Prepair
for
iece-parts
Factor of
Unit
module
cost Total
of
costspares repair
Active
time retrofit
of
Cost
and
Administration
repair
of
Cost modification
Facilities
16 MAINTAINABILITY ENGINEERING
19. The F/A-18 was developed under the Navy's "new look" weapons
procurement program in an effort to improve reliability and main
tainability and thus, reduce life-cycle cost [10]. During the four-
year period after entering service, the F/A-18 was three times
more reliable than the two aircraft it replaces, the F-4J and the
A-7E. The F/A-18 has an average of 2.2 mean flight hours be
tween failures (MFHBF) compared with its closest competi
tor the F-4J averaging 0.8 MFHBF. The F/A-18 required
26.0 MMH/FH which was a great improvement over the 46.1
MMH/FE necessary for the A-7E.
21. Pump failures cost an average of $4,000 per repair in the petro
chemical industry. The MTBF for typical pumps is 18 months
and because of the number of them at large refineries main
tenance costs for pump repairs alone may exceed $3,000,000.
Exxon has introduced a pump failure reduction program which
has yielded significant results and led to 29% less failures after
the first year of implementation [12].
dand
midual
c-geTABLE
between
oRic1.5
temhalpin/aorbcptailscioatlny fr/106
hr
Mechanical 15,000 - 1,830 - 4,370 - - 21,000 48
Dual
Dual
- -
xlO6
28.65
Digital/ Optical -71.608x10 _63.307x10 "334.68x10 ~42.537x10 5.5 X10-1 ~334.94x10
safety,
Flight
[13].
control
flight
systems fr/106
hr
Dual
cMoecnhtarnoilcasl
Hydromechanical
cForce
ontrol er Failure Elecpower
monitor trical
AFCS Total
*
18 APPLICATIONS AND BENEFITS
PROBLEMS
1-1. What is maintainability?
1-2. What is maintainability engineering?
1-3. What in today's advanced society dictates the acquisition and
the application of the maintainability engineering principles?
1-4. Name three complex products of today which should be de
signed by and their performance monitored through main
tainability engineering.
1-5. Name three current space and defense projects which have
been designed by and whose performance is being monitored
through maintainability engineering.
1-6. Which specific phases of existence of a product or system
does maintainability deal with?
1-7. List five of the most important applications of and benefits
derived from maintainability engineering in your opinion.
1-8. Illustrate by two numerical examples why product or sys
tem complexity dictates the acquisition and application of
the knowledge of maintainability engineering.
1-9. Why is today's worldwide industrial competition a challenge
to maintainability engineering?
1-10. How do you think maintainability specifications are set?
REFERENCES
1. Billit, A.B., "Control of Maintainability in Aerospace Fluid Power Sys
tems," Aerospace Reliability and Maintainability Conference Proceed
ings, Washington, D.C., pp. 340-349, June 1964.
2. Harter, W.W., "Results of an Airplane Reliability Program," Aerospace
Reliability and Maintainability Conference Proceedings, Washington,
D.C., pp. 65-70, June 1964.
3. Powell, H.R., "The Minuteman Approach to System Reliability," Aero
space and Maintainability Conference Proceedings, Washington, D.C.,
pp. 51-58, June 1964.
4. Camarata, J., "Product Reliability- The Concept of Integrated Relia
bility and Quality Assurance," Aerospace Reliability and Maintainabil
ity Conference, Washington, D.C., pp. 172-178, May 1963.
REFERENCES 19
2.1.1 WSEIAC
The concept of System Effectiveness as developed by the Weapons Sys
tem Effectiveness Industry Advisory Committee (WSEIAC), involved
the totality of the requirements associated with the system's avail
ability, dependability and capability, as well as reliability. It must be
pointed out that reliability is directly a component of availability and
of dependability. Furthermore, reliability has to be designed into a
system. It affects its design and in turn its performance or capability.
WSEIAC task groups prepared and published six reports in eleven vol
umes. Their titles and Defense Technical Information Center numbers
are as follows:
21
22 SYSTEM EFFECTIVENESS
2.1.2 ARINC
Another set of System Effectiveness concepts have been developed by
ARINC Research Corporation, 2552 Riva Rd., Annapolis, MD 21401.
A summary of these concepts is presented in Fig. 2.1. The definitions
of the System Effectiveness concepts are given in Table 2.1 and the
time categories involved are given in Table 2.2.
These efforts help to properly place reliability engineering in the
overall picture and in the concept of System Effectiveness. They point
out more vividly the importance of reliability engineering as an overall
concept encompassing reliability, maintainability, availability, opera
tional readiness, dependability, design adequacy and capability.
RELIABILITY ENGINEERING CONCEPT 23
System
effectiveness
Operating time
(Reliability)
Logistic time
Administrative
time
Fig. 2. 1 - System Effectiveness concepts.
24 SYSTEM EFFECTIVENESS
_ NT(t) - NF(t)
(2.2)
NT(t) '
therefore,
where
^g = unreliability, (2.4)
and
where
Nav = number of systems that are available to start
their mission successfully.
A quantitative definition of mission reliability, Rm, is
where
Ncm — number of systems that, having started their mis
sion successfully, complete their mission success-
fully.
Design adequacy is a difficult concept to conceive because it is dif
ficult to quantify. It can be illustrated by examples, however. Let
us assume one wants to take six people to work, there is only one
car available, and the car's capacity is only four passengers. The de
sign adequacy of this car for this mission is 4/6, or 67%. In terms
of intercontinental ballistic missiles, the probability that the required
number of missiles have been checked out and are ready to be launched
at the designated time is operational readiness. The probability that
each missile is successfully launched, reaches the target and explodes
successfully is mission reliability. The probability that the missile de
stroys the specified target to the extent intended is design adequacy.
If the number specified to destroy the whole target were launched suc
cessfully, and an inspection revealed that only 90% of the target was
destroyed, then the design adequacy of these missiles is 90%.
A quantitative definition of design adequacy, Da, is
D*-mt' (2-12)
RELIABILITY ENGINEERING CONCEPT 29
or
NAC 100
Nt = IF = ou = 119 systems-
3. The number of systems that will be operationally ready to start
their missions is obtained from
which yields
Nav = NT x Or = 119 x 0.98 =116 systems.
30 SYSTEM EFFECTIVENESS
D Ncm
Rm = -=j— ,
Nav
or
* Ncm'
or
22. Help promote sales on the basis of the reliability and maintain
ability of the products manufactured.
23. Provide a cost analysis technique whereby the optimum product
reliability and maintainability can be established at which the
total cost of the product to the customer is minimum. Total
cost is the sum of the initial cost or the purchase price, plus the
support cost or the cost of operating the machine, servicing and
maintaining it and the downtime cost for the designed-for life of
the product.
24. Increase the potential of the product as a defense or space prod
uct.
25. Reduce warranty cost, or for the same cost increase the length of
warranty.
34 SYSTEM EFFECTIVENESS
27. Establish the time required for life testing, and reliability and
maintainability demonstration tests.
31. Establish the time required for average system scheduled repairs.
37. Establish the overall man-hours required for the entire mainte
nance procedure with inspections.
38. Provide analyses of failure reports to see if all failures are of the
same type, if shipping and packaging methods are adequate, if
there are trends in the frequency of failure versus service life, if
there is a sufficient number of men available for maintenance, if
the downtimes and repair times are consistent with the estimates
and if the changes made affected the life and the maintainability
of the equipment and to the desired degree.
REFERENCES
1. Kececioglu, Dimitri B., Reliability Engineering Handbook, DEStech Pub
lications, 1148 Elizabeth Ave., #2, Lancaster, PA 17601-4359, Vol.1,
720 pp. 2002.
2. Kececioglu, Dimitri B., Reliability Engineering Handbook, DEStech Pub
lications, 1148 Elizabeth Ave., #2, Lancaster, PA 17601-4359, Vol.2,
568 pp., 2002.
3. Kececioglu, Dimitri B., Reliability & Life Testing Handbook, DEStech
Publications, 1148 Elizabeth Ave., #2, Lancaster, PA 17601-4359,
Vol.1, 960 pp., 2002.
4. Kececioglu, Dimitri B., Reliability & Life Testing Handbook, DEStech
Publications, 1148 Elizabeth Ave., #2, Lancaster, PA 17601-4359,
Vol.2, 900 pp., 2002.
Chapter 3
MAINTENANCE
37
SYSTEM SAFETY
safety.
amuvcycles
and of
Fig.
The
nia3.1
ritelainba-ility,
AVAILABILITY
\J~ SYSTEM
MAINTAINABIL TY
SYSTEM
C•OANSDIETISON
•FINDWEAROUTS
UNRELIABILITY
SYSTEM
GO
RESTORATION ACTION
TIME
LOGGING
FREE
TIME
CHECKOUT
AND
REPAIR
TIME
DELAY TIME REMOVE. REPLACE
V INACTIVE
TIME
TIME
DELAY
SUPPLY
FAULT
COR ECTION
TIME
MODIFICATION
DOWNTIME REPAIR PLACE
TIME IN
ADJUSTMENT- CALIBRATION
EQUIPMENT
TIME TIME of
life
the
with
eatimes
squoicpamtendt.
MAINTENANCE
COR ECTIVE
MAINTENANCE TIME
OVERHAUL
TIME OBTAINMINT MA|OR TIME
ITEM TIME
ACTIVE
TIME
OVERHAUL
MINOR TIME
ISOLATION
FAILURE
TIME
MISSION TIME
REPLACINC
AGINC UNITS TIME
PREPARATION
TIME
REACTION
TIME
(TURN
AROUND)
SERVICING
TIME
MAINTENANCE
PREVENTIVE
TIME
TIME
ALERT
INSPECTION
TIME
40 MAINTENANCE
l
42 MAINTENANCE
even though design provisions for minimizing such delays can some
times be made.
Most maintainability predictions are based on the distribution of
the active maintenance time and on the evaluated mean active main
tenance time of the equipment.
All of the above maintenance times can to a lesser or greater ex
tent be minimized or even eliminated by designing the right reliability,
maintainability, and safety into equipment and systems. Consequently,
desired reliability, maintainability and safety levels should be designed
into all equipment and systems.
12. Checkout time is the time required to test and observe the oper
ating characteristics and outputs of the equipment to determine
whether it can be put back to service again and function satis
factorily or within the specification requirements.
13. Clean up time is the time required to tidy up and clean up the
equipment and its immediate surroundings.
14. Logging the restoration action time is a necessary time that should
be devoted to documenting the whole active maintenance and
restorative action to insure that all reliability and maintainabil
ity data is properly documented for subsequent reduction to re
liability and maintainability parameters, for analysis, for design
reviews and for feedback to the disciplines responsible for the fail
ures and malfunctions. The objective is to improve the reliability
and maintainability of equipment and systems.
PROBLEMS
MAINTAINABILITY DESIGN
CRITERIA
49
50 MAINTAINABILITY DESIGN CRITERIA
13. Correctly number the parts according to circuit data flow and
their physical location.
14. Provide rests on which subassemblies can be placed to perform
the required maintenance without damaging their parts and com
ponents.
15. Be sure maintenance can be performed without endangering per
sonnel or system safety.
Figure 4.1 (a) illustrates one way of providing a more efficient,
tailored, completely automatic, built-in test unit which requires only
five minutes to diagnose the cause of failures of electronic equipment,
as opposed to using an auxiliary, general purpose test set-up, shown in
Fig. 4.1 (b), which requires 45 minutes to perform the same diagnostic
tasks.
Figure 4.2 (a) illustrates a better design which provides greater
accessibility for ease of trouble-shooting; removal and replacement of
failed, or malfunctioning, or worn-out units; as opposed to a design
which is hardly accessible, as shown in Fig. 4.2 (b).
Figure 4.3 shows outstanding design features of a single-stage steam
turbine which provides very good accessibility and ease of maintenance.
Figure 4.4 (a) gives the minimum openings for using common hand
tools to provide adequate accessibility, and Fig. 4.4 (b) gives the space
required for using these common hand tools to provide adequate ac
cessibility.
Figure 4.5 shows good and bad packaging practices, the preferred
one being that of a design for functional unitization corresponding to
separable modularization with minimum crossovers or interconnections
between the modules.
Figure 4.6 shows the benefits of standardizing parts, components
and subassemblies, and then designing them to be interchangeable so
that there will be fewer of them to procure, catalog, stock, inventory
and locate.
Figure 4.7 illustrates the need for good identification of parts and
test points, and for grouping of test points in an orderly, easily identi
fiable configuration.
Figure 4.8 (a) illustrates the need for coding each plug to its re
ceptacle so that maintenance errors through inadvertent interchange
of connectors are eliminated.
Figure 4.8 (b) shows the correct placement of the labels so that
they can be seen and read easily.
PROBLEMS 51
PROBLEMS
4-1. Come up with two suggestions of your own on how you would
improve the design of a certain assembly with respect to its com
ponents accessibility, adjustments, identification, safety, replace
ment, modularization, etc.
4-2. Consider your own car. Is there any system or subsystem that
could be improved with respect to its maintainability design cri
teria? What are these improvements? How would you implement
these improvements? Document your suggestions by appropriate
drawings.
4-3. Consider the failure of the water pump in your car. Come up with
your estimate of the expected time to disassemble it, replace it
and assemble it again including the minimum necessary time to
check it out to assure its proper function.
4-4. Obtain estimates of the repair time for the water pump in Prob
lem 4-3 from a car dealer or a car repair service.
4-5. How and where would you locate the water pump in your car for
ease of maintenance? Document your suggestions with appropri
ate drawings.
4-6. Where and how would you locate your car engine's oil dipstick
for ease of preventive maintenance thereof? Document your sug
gestions with appropriate drawings.
«
o
0
1) C
■r. o
EL 3
tr 3
3
a
e
u
-a
■i ■
C *—. -*—•
E 8
oo ,y 3
bi
s ■» o
a E >
X V~t
3 Tf
a , 8.
"3
| E C
u. «-*
u
Q.
T3
3 8
o ■—»
*■ Ul
UJ <w>
"*~0
C/i
o O "3
u
tn — oc a
^
--J
c* O U OX c
UJ —i o <~> H-
4-1
3
C
M
s
JO ■o
UJ to -J ^ u o
2 o UJ v>
tn
1
E
^
c
o CJ
3 K
:3 <-s
c
O 3 r3
U 9 •o
5. 5 h
5 >o.
o -»—a
■a 8 c
.1) "-1 1
1) « •f
l~ 0)
^^ tk
*—- LL.
52
trouble
of
for
aprovide
cdesigns
nto
ease
eoand
sPnrFig.
pi4.2
erbfileir-tyred
N(b)
oPn(a)
rpefer red.
rand
removal
eshooting,
placement.
u
54 MAINTAINABILITY DESIGN CRITERIA
/
Overall, a standard turbine that'ikasiaLta-maifllaml
DecaTJseTouTarT^eT^TTffS'parTs^no^asier^^^^
!
bO
B
i
to
s
C3
V
o.
O
S
3
•ae
a
«
to
55
hand
tools
provide
to
for
required
using
4.4(b)
cworking
Minimum
oFig.
space
m o-n
ac es ibility.
adequate
s
CI
—
LU —
0
r.
C
—
it
z
■- a
LU
4
g <aC
= 'Sb
to jg
C9
I
60
57
<
03
g
Q
O ?
O *5
CD w
/:
—j
ffl =
-
C
<
~ :/:
«J
**
y C
u
X '■—
~^ a
3Sa B
0
c u
> ■aa
*J
irz C
.-5 "3
s
si;
JJ
Ti
a s
— a
—J w
o — ci
9 u
5 a
_ i. ..
0 c
•J u
«J i
3
k
■o
n
■
—s §
—
c
T
M
h
58
GROUPING
OF
TEST
POINTS
GOOD
BAD
Kland
oper
eigrouping
of
tpoints
for
.tost
riof
amta.easeionitenance.
'•
•*•
•<•• •»•
»>• ••'
•!•••• •!•
JLLl
••
•C
!•
.1
••
••
.1
•..»
IDOF
ENTIFICATION
TEST
POINTS Fig.
4.7
Proper
iden-tific
CO
HERE
IN
NOT
HERE
UNDER
NOT
TOP)
(ON
HERE easily
he
sshould
eLabels
,.(b)
mof
aifor
nlabels
teof
easenapnd
clacement
e.
canoindenctP'°Per
of iof4-8
irFig.
csati"on
recede
the
*.
to
each
CW
ode
DAD attached.
beis
it
to
liich
GOOD
to
w
s
lugs
riral
elect
and
wiring
egrips
good
fFig.
of
with
P4.9
xarstoveirns-eiarolsning
THIS
HAS
DRILLED
BE
KIND
TO
DAMAGED
IS
SLOT
IF
OUT
GRIP
EXTERNAL
FOR
PMAKE
ROVISIONS
FREPAIRS
LUGS
U-TYPE
ACILITATE
BAD
(LUGS) mfwhich
aicnitleinatncte.
GRIP
EXTERNAL
PERMITS
HEAD
WRENCH
OF
USE
GOOD
^
damaged
main
during
would
sthat
get
unot esocmbalu-itesions for
bpSafety
Fig.
ars4.10 mplaced
abe
units
which
iproviding
nsuch
rests
tecanntonaasencnea. nce,
BAD tJODP
for
placed
mabe
units
which
iProvide
ntrests
ecannonance.
GOOD Uyuuu
KD
PROBLEMS 63
Cfla* CW
CT* 0«i4
CVt« Ciwl *pr
1 Cft-| u-i
This
c^a
Hi
omw |
oPf
off
ct-e
Not This
orr
tw
1 1 i i i i i
l-h pi r^ r4n i4n
t If t f f t t f
1 VWW 1 U- i 1
1
'8
•VWV—|h
CM —L '7
6/Ty oc
=3
6 ^> so
65
66 MAINTAINABILITY DESIGN CRITERIA
4-7. Where and how would you locate your car transmission's oil dip
stick for ease of preventive maintenance thereof? Document your
suggestions with appropriate drawings.
4-8. Where and how would you locate your car engine's spark plugs
for ease of maintenance thereof? What special tools if any, would
you use? Document your suggestions with appropriate drawings.
4-9. Come up with a better example than that given in Fig. 4.1 of
this chapter to decrease the diagnostic time of failed equipment.
Document your suggestions with appropriate drawings.
4-10. Come up with a better example than that given in Fig. 4.2
illustrating easier accessibility. Document your suggestions with
appropriate drawings.
Chapter 5
DOWNTIME DISTRIBUTIONS
AND MAINTAINABILITY
ENGINEERING FUNCTIONS
67
68 DOWNTIME DISTRIBUTIONS
V = ,=i (5.1)
1=1
EOl-r-W)7\2
Of - Ar - 1 (5.4)
70 DOWNTIME DISTRIBUTIONS
TV'
£ n^f - N{i'Y
ati = (5.5)
JV-1
7V'=29
£ »j = 46 = N, (5.6)
N'
£ njt'j = 30.30439, (5.7)
and
TV'
]T n,(«J)a = 75.84371. (5.8)
'75.84371 - 46(0.65879)2
av =
46- 1
or
at, = 1.11435.
Consequently, the lognormal pdf, representing the data in Table
DOWNTIME DISTRIBUTIONS 71
1 2 3 4 5 6
tj logetj - tj' wr ni TI& nAW
0.2 -1.60944 2.59029 1 -1.60944 2.59029
0.3 -1.20497 1.44935 1 -1.20397 1.44955
0.5 -0.69315 0.48045 4 -2.77260 1.92180
0.6 -0.51083 0.26094 2 -1.02166 0.52188
0.7 -0.35667 0.12721 3 -1.07001 0.38166
0.8 -0.22314 0.04979 2 -0.44628 0.09958
1.0 0.00000 0.00000 4 0.00000 0.00000
1.1 0.09531 0.00908 1 0.09531 0.00901
1.3 0.26236 0.06884 1 0.26236 0.06884
1.5 0.40547 0.16444 4 1.62188 0.65760
2.0 0.69315 0.48045 2 1.38630 0.96090
2.2 0.78846 0.62167 1 0.78846 0.62167
2.5 0.91629 0.83959 1 0.91629 0.83959
2.7 0.99325 0.98655 1 0.99325 0.98655
3.0 1.09861 1.20695 2 2.19722 2.41390
3.3 1.93920 1.42545 2 2.38784 2.85090
4.0 1.38629 1.92181 2 2.77258 3.84362
4.5 1.50408 2.26225 1.50408 2.26225
4.7 1.54756 2.39495 1.54756 2.39495
5.0 1.60944 2.59029 1.60944 2.59029
5.4 1.68640 2.84394 1.68640 2.84394
5.5 1.70475 2.90617 1.70475 2.90617
7.0 1.94591 3.78657 1.94591 3.78657
7.5 2.01490 4.05983 2.01490 4.05983
8.8 2.17475 4.72955 2.17475 4.72955
9.0 2.19722 4.82780 2.19722 4.82780
10.3 2.33214 5.43890 2.33140 5.43890
22.0 3.09104 9.55454 3.09104 9.55454
24.5 3.19867 10.23151 3.19867 10.23151
72 DOWNTIME DISTRIBUTIONS
5.1, is
5(0 = e »( v \ (5.9)
f <7t> V27T
or
1 l/l'-0. 65879 \2
O(t) = ==e 3* 1.11435 I
yK ' (1.11435) *VS
where t' = loge i.
The plot of this pdf is given in Fig. 5.1 in terms of the straight
times in hours. See Table 5.3 for the g(t) values used.
The pdf of the loge t which is that of a normal distribution, or of
the t' values, is
or
/j/x 1 _l/t'-0. 85879)2
flf(r) = ==e «l 1.11435 J .
; (1.11435) y/2r
This pcf/ is that of a normal distribution and is shown plotted in
Fig. 5.2. See Table 5.3 for the g(t') values used.
2. The mean time to restore the system when this part fails, t, is
obtained from
7 _ e0.65879+^(1.11435)2
or
t = 3.595 hr,
This compares with a mean of t = 3.609 hr, which would be
obtained using the straight t's and averaging them. The differ
ence is due to the fact that the former value assumes all data
are exactly lognormally distributed which is not the case, and
the latter gives only one estimate of a statistic, the arithmetic
mean, of the times to restore which would be an estimate of one
of the parameters, the mean of a normal distribution were the
data normally distributed. However, if it is known that the t's
come from a lognormally distributed population the t = 3.595 hr
is the value closest to the true mean.
Table
in
Fig.
5.1
given
data
t6.1
pdf
lognormal
the
of
Plot
imes-to-restore
hr
rto
t,
Time
estore,
hr
3.595
t=
Mean_
straight
the
of
hours.
in
terms
is
0.5582
hr
1.932
t
t==
Median
Mode„m
0.4T
w
74 DOWNTIME DISTRIBUTIONS
1 2 3
Times to Probability Probability
restore, density, density,
t, hr <?(<) g(t') = g(\oge t)
0.02 0.00398 0.00008
0.10 0.10480 0.01048
0.20 0.22552 0.04510
0.30 0.29510 0.08853
0.50 0.34300 0.17150
0.70 0.33770 0.23639
1.00 0.30060 0.30060
1.40 0.24524 0.34334
1.80 0.19849 0.35728
2.00 0.17892 0.35784
2.40 0.14638 0.35130
3.00 0.11039 0.33118
3.40 0.09260 0.31483
4.00 0.07232 0.28929
4.40 0.06195 0.27258
5.00 0.04976 0.24880
6.00 0.03556 0.21351
7.00 0.02625 0.18373
8.00 0.01985 0.15884
9.00 0.01534 0.13804
10.00 0.01206 0.12061
20.00 0.00199 0.03971
30.00 0.00058 0.01733
40.00 — 0.00888
80.00 — 0.00132
•At the mode, t = 0.5582 hr, g(t) = 0.34470 and g{f) = 0.19247.
At the median, i = 1.932 hr, g(t) = 0.18530 and g(i') = 0.35800.
100
Fig.
of
Plot
6.2
the
lognormal
pdf
the
tdata
given
Table
in
i5.1
rme-st-or-e
5.889'
0.634
1.932
to
Time
hr
t,
restore,
of
the
lt'.
of
terms
loget
ogi,
aorr=ithms
0.4T
Si
76 DOWNTIME DISTRIBUTIONS
or
1 = 1.932 hr.
This means that in a large sample of t's, half of the t's will have
values smaller than t, and the other half will have values greater
than t.
4. The time by which one-half of the restorations of such systems
will be completed is the median, or
t = 1.932 hr.
5. The most frequently occurring, or observed, time to restore such
systems is the mode of the pdf of the t's, i, and is given by
t = e'-"2'1 = t e-°2', (5.13)
2 _ 0.65879- (1.11435)2.
i c ,
consequently,
t = 0.5582 hr.
6. The standard deviation of the times to restore such systems is
given by
,2
at = t (eV - 1)', (5.14)
at = 3.59549 [e(111435>2 - l] ^ ,
or
at = 5.641 hr.
This compares with the standard deviation of the straight t's of
the raw data, assuming they are normally distributed, or
at = 4.945 hr.
The difference between these two values is due to the skewness
of the data which favors the lognormal distribution.
ANALOGOUS ENGINEERING FUNCTIONS 77
EXAMPLE 5-2
Prove that
M(i)= /'*(*)*.
Jo
tft)
[1
M(t)}
g(t)
=•-
Functions (*)]
[1
g(t)
fi(t)
M=-■
mAnalogous
reand
TABLE
a5.4
inlgtiaen-n-beaiblritnyg
Meanigtianeabirlintyg
rSfe
MO= f(T)
g(t).
of
idrthe
that
i.e.,
lis
neospvtcorarpoi*7
rbeltaucvimteoindtv;er
MO 1^7}
MO
=
<K0
marate.
inortenance
or rerate,
Repair
rate, storation
Item of
Pdf
times mto
aintain
resortore.
A(r).e-/>T)rfT
/(r)
=
functions.
gineering A/(T)
(T)-=[1-Q(T)]
Functions f(T)
X(T)
R(T)
=•
Renlgianbeilritnyg
ffi
Mr)=
f(T) A(T)
Pdf
times
of
fato
ilure.
Item Failure
rate.
number
Item 1 2
00
/
[1-M(t)]dt
MTTR
-y+
=
/
tg(t)dt
MTTR=
Functions ' g(t)dt
M(ti)=
J
Meanigtianeabirlitnyg
M(U)
P{t<U)
= MM-l-.'F*** mfmi
MTTR
J-l
M(ti)
"(««>-»-#!$
cot
bympliet.ion
Pof
robability Mean
time
to
maintenance maiorntain
Item
restore.
/
R(T)dT
MTBF=-r+
j
JT
MTBF=
{T)dT
I
' /(T)
Q(Ti)
dT= i-{^
0(rI)
i-H(r,)
=
Functions
Renlgianbeilritnyg g(r1)
p(r<r1)
= Jy
Q(T1)
l-e=
MTBF=T=m J-,
J-1
CTABLE
o5.4
ntin-ued.
Q(Ti)
Pof
robability
failure
Tj.
by
Mean
time
Item between failures.
number
Item 3 4
2
80 ANALOGOUS ENGINEERING FUNCTIONS
Jo
= - / ev dv,
Jo
where
/ fi(t) dt = v,
Jo
and
fi(t) = -dv/dt.
Then,
M(t) = -°v «
0'
or
EXAMPLE 5-3
Given the times- to- repair data of Example 5-1 determine the repair
rate function and plot it.
SOLUTION TO EXAMPLE 5-3
The repair rate function is given by
where
_ 1 / loge t— 0.65879 »2
g(t) = 7=e "~(
2* 1U135
1.11135 '
(1.11435) i y/2ir
when
Example
5-1
of
data
the
frepair
the
uof
rate
Plot
nFig.
5.3
ctio-n
5-3.
Example
for
dilsrepair
ottimes
gthe
rto
niarebourtmeadl, y
hr
Time
t,
restore,
to
6
5
4
82 ANALOGOUS ENGINEERING FUNCTIONS
and
.f log, 1-0.64878
The plot of fj.(t) for various values of / is given in Fig. 5.3 based on
the calculated results given in Table 5.5.
A sample calculation follows:
For t = 2 hr,
g{t' = loge0 = K*' = 0.69315),
<t>[z{t' = 0.69513)]
rti,/'0.69315-0.6S879>\
_ v\ 1.11435 )
1.11435
^(0.03083)
1.11435 '
or
/./i
<?(*' = loge 2n hr)
u n = tttt^
0.3988 = 0.35784.
1.11435
Therefore,
g(t = 2 hr) =
0.35784
or
g(t = 2 hr) = 0.17892.
The maintainability for t = 2 hr is
/*(<'=0.69315)
0(2) rf»,
-oo
oo
0.03083
= 0(2) dz,
/
or
M(t = 2 hr) = 0.5123.
84 DOWNTIME DISTRIBUTIONS
Consequently,
*-»■>- rifeSW
0.17892
1 - 0.5123'
or
PROBLEMS
i
88 DOWNTIME DISTRIBUTIONS
g(t) = == e 2 { it > .
Do the following:
(1) Write down the maintainability function for this equipment.
(2) Calculate the maintainability of this equipment for an al
lowable time to restore of 5 hr.
(3) Calculate the mean time to restore this equipment.
(4) Calculate the 50% confidence level time to restore.
(5) Calculate the time to restore for a maintainability of 95%.
5-6. An equipment which requires restoration to satisfactory function
when its components fail has a lognormal pdf with the following
parameters:
f = 5 and at> — 0.5.
Do the following:
(1) Determine the probability density function of the times to
restore the equipment and plot it.
(2) Determine the maintainability function and plot it.
(3) Determine the repair rate function and plot it.
(4) Calculate the 50% confidence level time to restore the equip
ment.
(5) Calculate the time to restore for a maintainability of 95%.
5-7. Given is the following times to restore distribution of an equip
ment which requires restoration to satisfactory function when its
components fail:
Do the following:
(1) Write down the maintainability function for this equipment.
(2) Calculate the maintainabib'ty of this equipment for an al
lowable time to restore of 5 hr.
PROBLEMS 89
Do the following:
(1) Write down the maintainability function for this equipment.
(2) Calculate the maintainability of this equipment for an al
lowable time to restore of 5 hr.
(3) Calculate the mean time to restore this equipment.
(4) Calculate the 50% confidence level time to restore.
(5) Calculate the time to restore for a maintainability of 95%.
5-10. An exponential equipment has a repair rate of 5 repairs per hour.
Do the following:
(1) Write the time-to-restore distribution equation in terms of
the repair rate.
(2) Write the time-to-restore distribution equation in terms of
the MTTR and plot it.
90 DOWNTIME DISTRIBUTIONS
91
£
8
=
=
S3
E
-=v
t
I
to
-3
rt
s
■5
c
I
60
92
EXPONENTIAL CASE 93
that part, in time fj, when that part fails or has to be replaced
preventively.
4. Maintainability comprises those characteristics (both qualitative
and quantitative) of materials, design and installations which
make it possible to meet operational objectives with a minimum
expenditure of maintenance effort (in terms of manpower, skill
levels, test equipment, technical data, operating and maintenance
manuals, maintenance support organization and facilities) under
operational environmental conditions in which scheduled and un
scheduled maintenance is performed.
5. Maintainability is the rapidity with which failures and malfunc
tions are diagnosed and corrected, or preventive maintenance is
completed and the equipment is successfully checked out. It is a
function of interacting variables including those of the design con
figuration of the equipment, of accessibility of frequently failing
or malfunctioning parts, or of parts scheduled for more frequent
preventive maintenance on the one hand and available facilities
and appropriate manpower on the other.
g(t)
yy ' = ^
n e""' = MTTR
, *„ e-M¥TR , v(6.2)'
where
/x = equipment corrective repair, replacement or restora
tion rate, in restorations per unit time; e.g., restora
tions per hour,
and
MTTR = -. (6.3)
hr
Rt,
time,
estoration
hr
0.693
ERT
0.347
=
0.900
8
96 MAINTAINABILITY AND ITS QUANTIFICATION
1 2
*1 M(h), %
0.24850 t 22.0
0.51080 t 40.0
0.69315 t (Median time to restore) 50.0
MTTR = t 63.2
tMltAXmOM = 2.3026 t 90.0
tMMAXs0M = 2.9957 t 95.0
loge a = - (6.7)
MTTR'
*,MMAX
MTTR = (6.8)
loge a '
and
'%^x = -MTTR x loge a. (6.9)
EXAMPLE 6 -1
It is specified that with probability M(t) = 1 — a = 0.90, or with
a = 0.10, the maintenance time for a specific equipment must not
exceed one hour; i.e., tiuMAX = 1 hr. Determine the MTTR that has
to be designed into this equipment.
REPAIR RATE 97
loge a
or about 26 minutes.
But if the predicted failure rates are not correct, the measured MTTR
may deviate significantly from the predicted value, even though the
individual repair times initially were well estimated. When designing
an equipment for maintainability, prediction techniques such as those
in MIL-HDBK-472 are used. An MTTR estimate of an exponentially
fading equipment is obtained from
£A,*<
MTTR = ^ , (6.11)
£A,
i=i
where
TV = total number of replaceable or repairable components,
MTTR = (6.12)
EXAMPLE 6-2
A system consists of three replaceable units which have the follow
ing MTBF's and replacement times:
Subassembly 1 : MTBFX = 1, 000 hr, t[ = 1.0 hr,
Wc = ^ (6.14)
t'=l
E/i
where
M = total number of different active preventive main
tenance actions undertaken,
where the terms Aj,/j,Imcj, and t\tpi are as defined earlier. In this
equation the same units must be used for the AJa and f'jS, and the
same time units for the values of tMci and Tm~.
EKT=*m, (6.19)
where of, is the variance around the mean of the natural logarithm of
the repair times.
ERT. It is given by Eq. (6.20) which is identical with Eq. (6.19); i.e.,
It can also be directly obtained from the estimate of the mean, rop, of
the natural logarithms of the mean repair times, *,-, or
N _
t=l
or
loge[l - M(t)}
t= -
4.61
5.99
9.21
139
hr
t,
restore,
to
Time
0.99 0.95
s
106 MAINTAINABILITY AND ITS QUANTIFICATION
1 2
M(f), t,
% hr
50 1.38629
90 4.60517
95 5.99146
99 9.21034
where LN stands for lognormal, N for normal, and <j> for the standard
normal distribution,
t' = loge t, (6.27)
t[ - V
*(*i) (6.28)
ov
LOGNORMAL CASE 107
and t' and at-, are given by Eqs. (5.1) and (5.2), and (5.3) and (5.4),
respectively. Entering the standard normal distribution probability
tables with z(t[) yields M(ti) quantitatively.
It is also desirable to determine the maintenance time by which
a specific percentage of the maintenance actions will be completed
satisfactorily. This is also the time *i_a for which the maintainability
is 1 — q, or
and
<Tti
or
*1_0 = e?+2<ti-<.)<T''. (6.34)
EXAMPLE 6-4
Given the times- to- restore data of Example 5-1, do the following:
1. Determine the maintainability function and plot it.
2. Determine the maintainability for this part if the maintenance
action needs to be completed within 5 hr.
3. Determine the maintainability for a 20 hr completion time.
4. Determine the time within which 90% of the maintenance actions
for such parts will be completed.
108 MAINTAINABILITY AND ITS QUANTIFICATION
1 2
Maintainability,
1-a «(«U)
0.80 0.8416
0.85 1.036
0.90 1.282
0.95 1.645
0.99 2.326
1. The maintainability function for the system, when this part fails,
M(t), from Eq. (6.26), is
z(t>)
M(t) = / ««) dz,
J—oo
where
t' - t'
2(0 = -—,
t' = loge t,
t' = 0.65879,
and
Of = 1.11435.
The quantified M(t) values are shown plotted in Fig. 6.4.
2. The maintainability for this system, for a maintenance time of
5 hr is
/z(logci>)
z(logc5)
4>{z) dz
-oo
Example
of
6-4.
the
fmPlot
for
times
repair
Fig.
a6.4
uto
intcati-niaobinlity
9
Time
restore,
hr
to
t,
5
4
3
6
2
5 jr n CO c CO c m2
§
1 10 MAINTAINABILITY AND ITS Q UANTIFICATION
With
and
, 1.6094-0.65879 „ „.,.
*{log. 5) = j-^ = 0.85310,
/u.sajiu
0.85310
<Hz) dz,
-oo
or
M (5 hr) = 0.8032, or 80.32%.
This means that there is about an 80% chance that the mainte
nance action on these systems, when this part fails, will last 5 hr
or less, or the probability of completing the maintenance action
satisfactorily within 5 hr is about 80%.
3. The maintainability for this system when this part fails, for a
maintenance time of 20 hr, is
where
loge 20 = 2.9957
and
1.11435
Therefore,
/2.0972
4>{z) dz,
-oo
or
M(20 hr) = 0.9820, or 98.20%.
"w-iffr^- (6-36)
Substitution of Eq. (6.36) into Eq. (6.35) yields the maintainability
U = 7 + 77 (2.302585)1//3, (6.39)
if M(h) = 95% then
U = 1 + «J (2.995732)1/^. (6.40)
If the times-to-restore distribution parameters are 7 = 0.5 hr, /? =
2.5 and 77 = 3.0 hr, the time to restore this equipment, t\, for a main
tainability of M(ti) = 95%, from Eq. (6.40), is
*1 = 0.5 + 3.0 (2.995732)1/25,
or
<i = 5.15 hr.
This means that 95% of the maintenance, or restorative, actions
will last 5.15 hr or less; or that if we allow 5.15 hr for maintenance
to be completed successfully, 95% of these maintenance actions will be
completed successfully within 5.15 hr.
WEIBULL CASE 113
EXAMPLE 6-5
Given is the following Weibull times-to-restore pdf of equipment
which requires restoration to satisfactory function when its components
fail:
, , 3.5 /i-0.25\2S
3.5 ft -0.25Y-5 f-o.»^.s
where t is in hours.
Do the following:
1. Write down the maintainability function for this equipment.
2. Calculate the maintainability of this equipment for an allowable
time-to-restore period of 5 hr.
3. Calculate the mean time to restore this equipment.
4. Calculate the median time to restore this equipment.
5. Calculate the time to restore for a maintainability of 95%.
SOLUTIONS TO EXAMPLE 6-5
1. The maintainability function for the equipment is given by
M(t) = 1 - e InSrJ .
or
M(t = 5 hr) = 0.83875, or 83.875%.
r [— + 1 j = T(1.2857) = 0.8998.
114 MAINTAINABILITY AND ITS Q UANTIFICATION
Therefore,
MTTR = 0.25 + 4.0 (0.8998),
or
MTTR = 3.84920 hi.
M(tx)= J g(t)dt,
where
g(t) = time-to-restore distribution, which needs to
be known ahead of time,
and
ti = restoration period for which the maintainability
is to be predicted.
PROBLEMS 115
EXAMPLE 6-6
Failures were simulated in identical equipment operating in identi
cal environments, and the same maintenance crew restored this equip
ment to satisfactory function. Out of the 120 restoration times that
were clocked, 115 lasted 35 minutes or less.
Determine the a posteriori maintainability of this equipment for a
required restoration time of 35 minutes.
SOLUTION TO EXAMPLE 6-6
The a posteriori maintainability is given by
— 115
M(h = 35 min) = — = 0.958,
or
M{h = 35 min) = 95.8%.
Consequently, the average maintainability that has been demonstrated
is 95.8%, and that on the average 95.8% of the restorative or mainte
nance actions lasted 35 minutes or less.
PROBLEMS
6-1. Given is the times-to- repair data of Table 6.5 for a specific part
in a system. Do the following:
116 MAINTAINABILITY AND ITS QUANTIFICATION
g(t)
*V ' = — .
t (1.5) y/2lr e" <^3-) .
Do the following:
118 MAINTAINABILITY AND ITS QUANTIFICATION
Do the following:
(1) Write down the maintainability function for this equipment.
(2) Calculate the maintainability of this equipment for an al
lowable time to restore of 5 hr.
(3) Calculate the mean time to restore this equipment.
(4) Calculate the 50% confidence level time to restore.
(5) Calculate the time to restore for a maintainability of 95%.
6-5. A system is made up of three exponentially failing subsystems
having the following mean times between failures, and mean
times to actively restore the system when a subsystem fails:
Subsystem 1 MTBFi = 800 hr, MTTRX = tMcl = 1.5 hr,
Subsystem 2 MTBF2 = 600 hr,AfTTR2 = Tm* = 1.8 hr,
and
Sybsystem 3 MTBF3 = 400 hr, MTTR3 = tMc3 = 2.0 hr,
Do the following:
(1) Calculate the mean time to actively and correctively restore
the system when any one of the subsystems fails.
(2) If Subsystem 1 is subjected to three preventive maintenance
tasks, Subsystem 2 to two and Subsystem 3 to three, with
the frequencies and active times given in Table 6.7, calculate
the mean active preventive maintenance time of this system.
(3) Calculate the mean active corrective and preventive main
tenance time of this system.
120 MAINTAINABILITY AND ITS QUANTIFICATION
Do the following:
(1) Write down the maintainability function for this equipment.
(2) Calculate the maintainability of this equipment for an al
lowable time to restore of 5 hr.
(3) Calculate the mean time to restore this equipment.
(4) Calculate the 50% confidence level time to restore.
(5) Calculate the time to restore for a maintainability of 95%.
6-8. Given is the following time-to-restore distribution of an equip
ment which requires restoration to satisfactory function when its
components fail:
Do the following:
(1) Write down the maintainability function for this equipment.
(2) Calculate the maintainability of this equipment for an al
lowable time to restore of 5 hr.
PROBLEMS 121
— t _ t. t
tn = ti H t-2 H 1 tN,
m\ mi myv
123
124 MEAN TIME TO ACTIVELY RESTORE AN EQUIPMENT
or
N T
«=-«£(=).1=1
<T-l>
where
tR = mean time to correctively restore the downed equip
ment to successful function for t hours of operation,
»=i
and
consumed for this effort is d\ for Unit 1, e^ for Unit 2, etc., then Eq.
(7.1) becomes
(7.5)
where
t* = mean time to replace a failed unit with a fresh
one.
time,
rt0.
equipment
iesntmeanoraitnisoinc
the
and
of
g2(t)
2
Crew
Crew
of
g,(t)
1
Restoration
t
time,
g(t)
s
RESTORATION EFFICIENCY AND CONSISTENCY 127
EXAMPLE 7-1
It has been established that an equipment's intrinsic mean restora
tion time clocked for a select, skilled crew is 36 minutes. Two other
crews of the equipment manufacturer's Service Department are selected
to restore the equipment shipped to a key customer. The times to re
store this equipment by each crew are determined with the following
results:
Crew 1: t\ = 42 min and o\ — 4 min.
Crew 2: ti = 48 min and a2 = 8 min.
128 MEAN TIME TO ACTIVELY RESTORE AN EQUIPMENT
3. Which crew would you select from the restoration time efficiency
point of view?
4. Which crew would you select from the restoration time consis
tency point of view?
Et2 = to I h = ^ = 0.750.
C1 = l-a1/?T=l-^ = 0.905.
— t _ t — t —
Dr= — *i na H t2 n-i + (7.8)
mi m? tun
or
^=<te) (7.9)
t=i
35- M?)-1=1
(7.10)
If each unit has a constant failure rate, then the mean maintenance
man-hours for t cumulative operating hours is given by
N
I
Dr = t J2(Xi U m),
t'=l
(7.11)
' N
DR = E(A- E »»•)• (7.12)
«=1
also known as
D^ = MMH/OH, (7.13)
This value may be used to calculate the average labor cost, Cl, for
repairing and replacing failed equipment on the spot and/or in a repair
shop for t cumulative hours of operation of the equipment from
CL = DRR ■ cL, (7.16)
where
d = average labor cost to repair and replace, $, for t
operating hours of the equipment,
and
C£, = average labor cost per hour, or the average hourly
labor rate.
EXAMPLE 7-2
A subsystem consists of three units, which are reliabilitywise in
series. Their useful life failure rates and wear-out pdf parameters are
given in Table 7.1.
MEAN MAINTENANCE MAN-HOURS 131
1 2 3 4 5
Unit m r,
hr hr
1 0.50 l 0.75 2
2 0.80 3 1.00 3
3 0.25 2 0.50 2
1. What is the mean time to replace these three units, per subsys
tem operating hour?
2. What is the mean time to repair these three units, per subsystem
operating hour?
3. What is the mean time to repair and replace these three units,
per subsystem operating hour?
4. Same as Case 1 but for operating hours equal to the stabilized
MTDF of this subsystem with units reliabilitywise in series,
given by
132 MEAN MAINTENANCE MAN-HOURS
MTBFreg= r R,s(T)dT,
Jo
where Ra, (T) is the reliability function for the configuration and
the types of units comprising the configuration.
8. Same as Case 2 but for operating hours equal to the regular
statistical MTBF of this subsystem.
9. Same as Case 3 but for operating hours equal to the regular
statistical MTBF of this subsystem.
10. What is the average total maintenance man-hours to repair and
replace, per subsystem operating hour, if the failed units are
not disposable but require repair in a workshop before they are
replaced?
11. What will the results of Case 10 be if Units 1 and 3 can be
repaired and replaced simultaneously, but by the two separate
crews used in Case 10?
12. What will be the mean time to repair and replace these three
units, per subsystem operating hour, for Case 11?
MEAN MAINTENANCE MAN-HOURS 133
1. The mean time to replace these three units, per subsystem oper
ating hour, is determined from
3 /*
£? Vm«
The MTBF for each unit is calculated from
m; = r Ri(T) dT.
Jo
For Unit 1
™i = r Ri{T)dT
Jo
= J" (o.25 e-00003 T + 0.75 e~lM 5) dT,
1
mi = 0.25 + 0.75
0.0003
0-25 + 0-75
n „ o + 300 r
mi = nnnM
0.0003 AtOl-
where
T f — + l) = r(1.4) = 0.88726.
Therefore,
0.25
mi = + 0.75 [(300)(0.88726)] = 1,032.96683 hr.
0.0003
Similarly,
T7l2 = / R2(T) dT
Jo
= J°° (o.25 e"00006 T + 0.75 e^]3 ') dT,
0.25
™2 = 7-^77 + 0.75
0.0006
0 + 500 r
(1-M
134 MEAN MAINTENANCE MAN-HOURS
where
I (rr + l ) - I'D. ■2^7) 0.S9975.
therefore.
0 25
m2 = - ' + 0-75 [(500) (0.89975)] = 754.07292 hr;
and
roc
m3 = / R3(T) dT
Jo
= |o°° (o.25 e-000045 T + 0.75 e^]' *) dT,
0-25
77l33 = 0.00045 1- 0.75 u ! .Kiorf JL + i
4.5
where
T (— + l) = r(1.22) = 0.91257.
Therefore,
0 25
m3 = nn'nAe + °-75 [(400)(0.91257)] = 829.32656 hr.
0.00045
Substituting these values into the equation for V, yields
t*\ 0.50 0.80 0.25
"■fiO-i:
032.96683 754.07292 829.32656'
2. The mean time to repair these three units, per subsystem oper
ating hour when all units require repair, is determined from
3. The mean time to repair and replace these three units, per sub
system operating hour, is given by
*-££?)•
__ (0.75 + 0.50) (1.00 + 0.80) (0.50 + 0.25)
1,032.96683 754.07292 829.32656 '
or
^RR ~ 0.004501492 hr/subsystem operating hr.
4. The mean time to replace these three units for operating hours
equal to the stabilized MTBF of this subsystem
t = m„ • i",
where
1
msa = subsystem MTBF =
E—
1=1
m.. =
J + I +
1,032.96683 T 754.07292 ' 829.323656
L
or
m„ = 285.71309 hr.
Therefore,
I = (285.71309)(0.001846397),
or
I = 0.527539796 hr.
5. The mean time to repair these three units for operating hours
equal to the stabilized MTBF of this subsystem is given by
d = mM • dm,
= (285.71309)(0.002655094),
or
2 = 0.758595111 hr.
136 MEAN MAINTENANCE MAN-HOURS
6. The mean time to repair and replace these three units for oper
ating hours equal to the stabilized MTBF of this subsystem is
given by
dRR = m„ • dRR,
= (285.71309)(0.004501492),
or
dRR = 1.286135189 hr.
7. The mean time to replace these three units for operating hours
equal to the regular, statistical MTBF of this subsystem is
I = MTBFreg ■ F,
where
MTBFREG = /°° R..(T) dT,
Jo
= r°s,(T).jfe(T).Jk(r)<Br,
= /°° [ (0.25 e-00003 T + 0.75 e~^ *)
.(0.25e-OOOO6T + 0.75e-[^o]35)
• (0.25 e-000045 T + 0.75 c"I*J") ] dT.
8. The mean time to repair these three units for operating hours
equal to the regular, statistical MTBF of this subsystem is
1 = MTBFreg -F,
= (260)(0.002655094),
or
d = 0.690324440 hr.
MEAN MAINTENANCE MAN-HOURS 137
9. The mean time to repair and replace these three units for oper
ating hours equal to the regular, statistical MTBF of this sub
system is
dRR = MTBFreg • dfm,
= (260)(0.004501492),
or
dRR = 1.170387920 hr.
TE.f./5aiia"
i=i
11. The DRR of Case 10, if Units 1 and 3 can be repaired and re
placed simultaneously, but by the two separate crews, is the same
since the replacing and repairing man-hours are independent of
simultaneity, as long as these two crews are putting in their time
anyway. Thus, even though the repair and replacement of Units
1 and 3 occur simultaneously, the mean man-hours per operating
hour remain unchanged.
12. The mean time to repair and replace these three units, per sub
system operating hour, for Case 11 is given by
^-tm i=i
138 MEAN TIME FOR MAINTENANCE
where
A'
H(A« Q + Api dpi-) = (289 x 10"6) (0.55) + (310 x 10~6) (0.75)
i=i
PROBLEMS
and
r3 = 2.
!
142 MEAN TIME FOR MAINTENANCE
m.. =
N i '
E—
:=1
Jo
e"^ i I dT = f + »rg+i).
or use Simpson's Rule with 8 or 16 intervals. Note that the
regular, statistical MTBF for any configuration is given by
MTBFreg •f Jo
Raa dT,
where Rss is the reliability function for the configuration and the
types of units comprising the configuration.
7-2. A system consists of three units which are reliabilitywise in series.
Their useful life failure rates and wear-out probability density
function parameters are given in Table 7.5.
TABLE 7.5 - Useful life and wear-out data associated
with the three units of the subsystem
for Problem 7-2.
and
£ = 0.33.
Assume the stabilized MTBF of each unit is given by
MTBFi = l
*ci T t ■
1. What is the mean time to replace these three units per system
operating hour if the average replacement times, i*, are as
follows:
i\ = 0.55 hr,
t'2 = 0.75 hr,
and
0.33.
Furthermore, the required number of service men, n,, are
the following:
m = i,
n2 = 3,
and
n3 = 2.
2. Same as Case 1 but for operating hours equal to the stabilized
MTBF of the units.
3. What is the average total maintenance- man hours to repair
and replace per system operating hour, if the failed units
are not disposable but require repair in a workshop before
PROBLEMS 145
they are replaced and the average repair times, </,-, are as
follows:
dx = 1.5,
d2 = 2.0
and
J3 = 1.0.
The number of required repairmen, r,, are as follows:
n = 2,
r2 = 3,
and
r3 = 2.
4. What will be the results of Case 3 if parts 1 and 3 can be
repaired and replaced simultaneously?
Note that the stabilized MTBF for a system with units reliabil-
itywise in series is given by
1
m.. = N
£
Also note that
MAINTAINABILITY
ENGINEERING
SPECIFICATIONS AND THEIR
APPLICATIONS
147
148 MAINTAINABILITY ENGINEERING SPECIFICATIONS
8.3.1 REQUIREMENTS
Maintainability verification, demonstration, and evaluation are required
to be performed in accordance with the maintainability, M, test plan
prepared by the contractor and approved by the procuring activity.
The M test plan shall be totally responsive to the qualitative and
quantitative requirements and supplemental information contained in
the procurement documents and the M program plan required by MIL
- STD - 470. The plan shall embody three phases at the system level:
1. Phase I - Verification.
2. Phase II - Demonstration.
1. Choose
Ho = Ho = Ili = lower, one-sided confidence limit on
the mean time to restore the failed equipment,
Hi — p.\ — tu\ — upper, one-sided confidence limit on
the mean time to restore the failed equipment,
f3 = consumer's risk,
a = producer's risk,
and
Plan A, which assumes a lognormal g(t),
or
Plan B, for which selection of a g(t) is not necessary.
Figure 8.1 illustrates these parameters as they relate to the mean-
time-to-restore distribution, g(t); and Fig. 8.2 as they relate to
the Operating Characteristic (OC) Curve.
2. Calculate the demonstration sample size, n, for Test Plan A,
from
popolauon ejlocatioa
Drfflt
It At E-l 3 10 3 Ho
(i«) •4 I*
B-7 3« D-4
C-7
Relative of
Iraq,
173
0 OK
0 70
42 0.0(4 0.047 0.013 166
bo) 6l»7
Tolol
m rata" 163 loa 10 240 30
2*
• 
pStTABLE
r8.1
aotcifei-cdautirone. groepief
Tul Gro.p TaaaA Group
1 2
A, B.C.D.E
Taaae
3
Group
r Oroup
Tuii 0
Tut
A
Group
3
B
Tut
Group
0
AT
Tut Group
Tub
A
m
of
each
«Um«nl
limo
(iaclude
perform
lime
rlo
eR/R
movem/replace
Dole cjrcl*
1.0 1.0 0
> 10 10 1.0 1.0 or 0.7 •
0 I
0
I*. ill
io
ronly
to
ni*
labia
(or
oliioipThi.
eari■•lodralodiela
M*ial«n*nc«
Cr».«al.(B)
<M •«•• Replace
R/R(A) R/R(A) R/R(B) R/R(C)R/R(D) R/R(E) R/R<F) R/R(A) R/R(A) R/R<A)
MIdofiaed
i.
L-Su
TD-731)
supply
Powti Transmit er trackar
Prcq
F«l«li0«l
level
or Modulator Radar angle indicator
«el control Drift
XYZ
Radar
Kquip•rncnt Amplifier
<»)
IF-
A IF-B
unit*
Major angle indicator
frequency
tracker act control Out*
Rada*
(l) NOTE
2: NOTE
3 NOTE
4
Cn
152 MAINTAINABILITY ENGINEERING SPECIFICATIONS
>—-
a = producer's
risk
P(A)=l-a
<
■— 1-P=P(R)
n= Ml -MO
where
/3 = consumer's risk,
or
P(H > /zi = hi) < 0,
a = producer's risk,
or
X < /iQ + 2C
v^T
154 MAINTAINABILITY ENGINEERING SPECIFICATIONS
NOTE: This table is for illustration only and does not reflect real failure modes, etc.
otherwise reject.
For Test B
Accept if
X < no + za
y/n'
otherwise reject.
Here
n
x = '^— = t,
n
and
£ t2i - n t2
d* = a? = & — = fft\
n- 1
EXAMPLE 8-1
The maximum acceptable mean maintenance time is
Mi = tvi = 55 minutes,
TEST METHOD 1 155
P$>'kn)<0= io%.
The desirable, design-to, mean maintenance time is
Mo = tn = 35 minutes,
with a producer's risk, a, of 5%, such that
P(?<*Li)<<* = 5%.
Find the number of the maintenance actions that have to be demon
strated, n, according to Plan A, and according to Plan B.
SOLUTIONS TO EXAMPLE 8-1
Plan A sample size
\2
(Ml - Mo)2 V *
or
n = 34,
with
o\i = 0.6, a prior estimate.
n=
d
Then,
/l.65+1.28\2 nnnn
*=[ 55-35 J =30-99'
or
n = 31.
EXAMPLE 8-2
If t = 40 minutes and at — 30 minutes, determine if the equipment
passes its maintainability demonstration test according to Plan A, and
according to Plan B.
SOLUTIONS TO EXAMPLE 8-2
Plan A
o d
t < MO + Za-7=,
y/n
y/n
or
c 30
t < 35+ 1.65-^= = 43.49.
V34
Since
40 < 43.49,
the equipment passes the test.
Plan B
- _ d
or
40 < 35+1.65-^L = 43.89.
Since
40 < 43.89,
the equipment passes the test.
TEST METHOD 1 157
A realistic estimate of t', the true mean of the times to restore the
equipment, is needed. One way is to start with t = no = 35 minutes
and increase it as experience indicates.
We know from the lognormal pd/'s properties that
logJ = F+-at2„
or
t' = logJ-^,.
F = log, 35 -^ (0.6),
or
t' = 3.255.
d2 = 1,006.3,
or
d = 31.7 minutes.
4. 0 — consumer's risk.
5. a = producer's risk.
The two times-to-restore distributions for the same chosen, fixed
maintainability are shown in Fig. 8.3. The corresponding OC curve is
shown in Fig. 8.4.
^2
where 5f,, is a prior estimate of the variance of the logarithms of the
times to restore.
2
o
H (N
II II
• =1 a
J w°
6
0)
E
ul (31*
J3 ct3
s«o a.
E- - Ol
ii n -o
o
*r «-? ,S
ii V
s
«->
1
A
&
159
160 MAINTAINABILITY ENGINEERING SPECIFICATIONS
. I>get,
r - i =i
n
and
1/2
'£(loget,)2-n<'2"
i=i
t' —
n-1
_
The decision criterion is
1/2
*'* = X' = loge tL + za ov -1 + * (1-M)
n 2(n-l)
Accept if
' + *(l-Af) "f S * ■
TEST METHOD 2 161
Reject otherwise.
EXAMPLE 8-3
It is specified that for a maintainability of 95% the desired mainte
nance time of the equipment be 35 minutes, and the 95% maintenance
time is not to exceed 55 minutes.
The consumer's risk is specified as 10%, and the producer's risk as
5%.
Determine the maintenance actions that have to be demonstrated,
assuming Bt> = 1.2.
SOLUTION TO EXAMPLE 8-3 i
Given are
a = 0.05, 0 = 0.10,
M = 95%, 100% -M = 5%,
tl = 35 minutes,
ty = 55 minutes.
Za = 20.05)^/9 = ^o.lO,
za = 1.65,2/3= 1.28,
t'u = loge 55 = 4.00733,
and
t'L = loge 35 = 3.55535.
Then,
2 + (1.65)2
n = (1.2)a 1.65 + 1.28 : \2 = 1jnoo
142.88,
4.00733-3.55535,
or
n = 143.
EXAMPLE 8-4
The maintenance actions of Example 8-3 are demonstrated with
the following results:
1 (1.65)2 1/2
Since
5.36 ft 3.81,
the equipment has not met the ti = 35 minutes requirement.
_1_
p
ML
T
I
;
T.
time
rfor
to
maeiTwo
Fig.
n8.5-
sonetatinoabrileti,es
M„
T
I
a;
W)
2
TEST METHOD 3 165
<
a.
and
[B] = za[Mu{\ - Mu)}1'2 + zp[ML(l - Mrf]1'2.
and
£ e-n(i-Mt) . [n(l-ML]r ^ ^^
r=0
Table 8.3 provides sampling plans for various a and /? risks and
ratios of
fc = Pj/Po = 1-^L when My > 0.80.
1 - Mu
Sampling
for
plans
TABLE
specified
(3
and
8.3-po,pi,a 31.800 8.810 5.580 3.080 2.300 1.540 0.824
D
0.05
a the
greatest
Uie
fi c 38 12 7 4 3 2 1
vD
by
te
&loe
p k »nd
0.05
■
33.000 10.300 5.430 3.150 1.750 1.100 0.532
D
0.10
m fi
a 0.10
m
c 40 14 8 5 3 2 1
ippropri*
the
ie
43.000 12.800 7.020 4.880 2.430 1.750 0.532
0.05=
D
0.10,4
0.10,a
=
P
c 51 17 10 7 4 3 1
4.
■
divi
0,
ind
a1, umber
ifc
30.200 8.250 4.700 3.280 1.870 1.370 0.353
D
p
0.20
m
n
ince
c 38 14 3 8 4 3 1
given
for
prj.p, he
accept
43.400 12.400 8.170 3.080 2.810 1.370 0.818
COS
m
a
T>
0.10
p= T
48.
c 54 10 7 5 3 2
1*
kite,
pie
r
the
nquo
fi
OmS D
0
is.too 8.480 5.430 3.280 1.870 0.818 -HI-
S4.100
lei»rr
cH 22 13 ■ 8 4 2 th«it
t&
I le
find
To integer
Po 1.5 2.0 2.5 3.0 4.0 SO 10.0 Then
n=
168 MAINTAINABILITY ENGINEERING SPECIFICATIONS
1. Mu = 0.90, then
P0 = 1 - Mv = 1 - 0.90 = 0.10.
Since P0 < 0.20, Table 8.3 is used to find c and n as follows:
a = 0.05, 3 = 0.10,
and
A _ 1-ML _. 1-0-70 = CU
P0 1-Mu 1-0.90 0.1
Then, from Table 8.3
D b 3.98;
consequently,
D D 3.98
' P0 I- Mv 1-0.90'
or
n = 39.8.
Use the greater integer value less than the quotient, then n = 39.
Therefore, 39 maintenance actions have to be demonstrated.
2. From Table 8.3 with a = 0.05, Q = 0.10, and k = 3,
c = 7.
3. Accept if r < c. Here,
r = 5 < c = 7;
consequently, the equipment should be accepted as having demon
strated a maintainability of at least M — 70% with 3 < 10%, for
a critical, desired maintenance time of T = 20 minutes.
TEST METHOD 4 169
4. In this case
r = (0.25) n,
or
r = (0.25)(39) = 9.75, or r = 10.
.£o°gio<«-)
log10 MTTRa = — = t" = log10 1, (8.2)
and
EOog10ic,)2 - nc(\og10MTTRG)^
t=i
S = Ofi = (8.3)
IL. - 1
where log10f = t", and MTTRq is the measured geometric mean time
to repair, or tc.
A.
170 MAINTAINABILITY ENGINEERING SPECIFICATIONS
8.7.4 DISCUSSION
The value of the equipment repair time, ERT, to be specified in the
detailed equipment specification should be determined from:
ERT( specified) = 0.37 ERTmax, (8.5)
where ERTmax is the maximum value of ERT that should be accepted
no more than 10% of the time, and 0.37 is a value resulting from the
application of the "Student's t" operating characteristics that assures
a 95% probability that an equipment having an acceptable ERT will
not be rejected as a result of the maintainability test when the sample
size is 20, and assuming a population standard deviation at» of 0.55.
EXAMPLE 8-7
A specific equipment's median repair time, ERT, requirement is
2.50 hours. Twenty (20) corrective maintenance tasks are performed,
in accordance with the procedure outlined in Appendix A of MIL-STD-
471, or Table 8.1, and their duration is recorded. Using these times to
repair, the following are calculated:
log10 MTTRo = 0.65,
where
log10 MTTRg = t".
Also
F7 = logi0{,
as
TEST METHOD 5 171
t = 10' .
It may be seen that MTTRq is the median time to restore the equip
ment to satisfactory function. The standard deviation of the twenty
(20) corrective maintenance actions is calculated to be <rt« = 5 = 0.85.
Determine if the equipment is meeting its repair time, ERT, require
ment.
SOLUTION TO EXAMPLE 8-7
Accept the equipment as having met its ERT requirement if
log10 MTTRG < log10 ERT + 0.397<rt»,
0.65 < log10 2.50 + 0.397(0.85),
0.65 < 0.3979 + 0.3375,
or
0.65 < 0.7354;
therefore, this equipment is meeting its ERT requirement.
(8.7)
EXAMPLE 8-8
An aircraft has a required operational readiness of 75%, with an
average daily usage of 2 hours per day and an average flight length of 4
hours per flight. The nonchargeable maintenance downtime per flight
is 0.2 hours per flight and the delay downtime is 1 hour per flight. Find
the chargeable maintenance downtime per flight, cj!f^p ■
£f£
a/y/n
-mi).
Equating these to their percentiles yields
A'0 - mo _
, i— — za<
a J\Jn
TEST METHOD 5 175
and
Xa -fii ■20.
a/y/n
Then, the test sample size can be shown to be
n= T (8.8)
(«?*)
If n < 50. then a sample of 50 shall be used.
EXAMPLE 8-9
For a requirement of /io = 2.0 hours, the following parameters were
agreed upon by the procuring agency and the contractor:
a = 0.10,
0 = 0.10,
/ii - no — 0.3 hours.
and
a — 1 hour.
Find the required number, n, for maintainability demonstration.
SOLUTION TO EXAMPLE 8-9
Since a = 0 = 0.10,
zQ = 1.28,
and
z3 = 1.28.
Then, from Eq. (8.8),
(1.28 + 1.28)2 6.57
n = (0-3/1)2 = 72.8,
0.09
or
n = 73.
NOF _
E tXi - *)2
t=i
s =
"N NOF - 1 '
or
\Of
s = J] X,2 - (NOF) -D-2
X (8.10)
V (NOF- 1) l i=i
Flight CMDT,
number hr
1 3.0
2 12.0
3 1.5
4 5.0
5 3.0
6 6.0
7 8.5
8 2.0
TEST METHOD 5 177
1 r 8
=
E
\ (8-1) .t=i
Y2 - (8) (5.125)2 i
where
= \/l3.0536,
or
5 = 3.6130 hours.
Since a = 0.10, za = 1.28, and the test for the decision procedure
is
zQ s (1.28) (3.6130)
Ho + 5+
y/NOF " ' y/$
= 5+1.6351,
= 6.6351, or 6.64/ir/ flight.
178 MAINTAINABILITY ENGINEERING SPECIFICATIONS
Then, since
X= 5.13 < 6.64,
the aircraft system has met its target CMDT requirement.
***•%$. ("«
where
CM M = total chargeable maintenance man-hours,
and
DFH = total demonstrated flight hours.
If the demonstrated MH R value is less than or equal to the re
quired man-hour rate, MH Rq, plus a value of maximum difference,
AMR, then the requirement is said to have been met. The value of
AM R is provided by the procuring agency, usually as a percentage of
the required system man-hour rate. This value is not based on the de
termination of quantitative producer's and consumer's risk levels, but
is based 011 prior experience with similar systems.
The decision procedure is as follows:
Accept if
MHR< MHR0 + AMR. (8.12)
Reject if
MHR> MHRo + AMR. (8.13)
EXAMPLE 8-11
During testing of a new aircraft system, a total of 1.500 flight hours
are accumulated. Also, 7,560 hours of chargeable maintenance down
time are incurred during the testing. If the required maintenance man-
hours for the system is 5 man-hours per flight hour, with a AMR of
10%, has the aircraft met the man-hour rate requirement?
TEST METHOD 7 179
Calculate AMR by
CMM
MHR
DFH '
7,560
1,500'
or
MHR = 5.04 man-hours/flight hour.
are calculated.
Table 8.1 is used to determine the total equipment failure rate, Aj,
and the MTBF from 1/AT.
The specified preventive maintenance tasks that will be required to
be performed in operating time,
T = n- (MTBF)
TEST METHOD 7 181
tXa + PS j
MHR = Xn-{MTBF) = MTBF {*c + ITJ '
4< = (£) • I
If Xc is normally distributed, it can be shown that
MTBF \Xc+ n)
variMiee = \ \mTbf) ■
Consequently,
„„n . 1 A» PS\ 1 d
MHR = ** * MTSF (Xc + Vj " Za7^MTBF-
Rearranging yields
EXAMPLE 8-12
In a specific equipment with a designed-in MTBF of 200 hours, 30
corrective maintenance actions, sampled per Table 8.1, are simulated.
The corrective maintenance man-hours expended for each action are
recorded. The average man-hours per corrective maintenance task is
calculated to be
30
_ £ xci
Xc = — = 2.5 man-hours,
30
and the standard deviation of the man-hours expended for the correc
tive maintenance tasks is calculated to be
d = 0.55 man-hours.
It is estimated that in a time period of
T = n- {MTBF) = 30 • (200),
or
T = 6, 000 hours,
the average total man-hours that will be required for all preventive
(scheduled) maintenance tasks would be
PS = 85 man-hours.
The producer's risk is chosen to be a = 5%.
Determine if this equipment has met its man-hour rate, MHR,
requirement of MHR = 0.050 man-hours per operating hour.
SOLUTION TO EXAMPLE 8-12
The decision criterion for Test Method 7 is the following:
If
PS d
Xc < MHR ■ (MTBF) + za-=,
n y/n
accept the equipment.
TEST METHOD 8 183
In this case
85 0 55
2.5 < 0.050(200)- — + 1-645-^=,
2.5 < 10-2.833 + 0.1652,
or
2.5 < 7.332.
Consequently, the equipment is accepted because it has met its MHR
requirement.
Here
HA) . . ,. -
k = = proportion ol repair times exceeding t,
PLAN Bx
rNumber
equipment
dof
emsotornosrtations, performed
cmnumber
of
actions
oarintecntainvce
Fig.
Plan
8.7-
Alt
Test
Method
8.
100
performed
mNumber
tasks
be
of
ato
intenance
60
80
40
20
Fig.
Plans
8.8-
Bt
and
B2,
Test
Method
8.
•2^•
s1
§ 8 x<u M 53
--S
U •as
00 CO
190 MAINTAINABILITY ENGINEERING SPECIFICATIONS
PLAN B2
Here
k - r^
or
* and tMmax=95% .
Choose the sample tasks by preparing Table 8.1 and basing Column
12 on a total sample size of 100. Choose variable sampling in conjunc
tion with a random number table uniformly distributed between 0 and
1, as per procedure of Column 13. Total up the maintenance tasks with
a duration exceeding the required values oft and <Mmoi- Compare these
totals with those given in Tables 8.4, 8.5 and 8.6, as applicable to the
two maintenance time requirements, for an accept or reject decision.
EXAMPLE 8-13
It is specified that the maintainability of the equipment be demon
strated on the following dual requirement basis:
1. A mean time to restore of t — 30 minutes.
2. A time for a 95% maintainability or for a <Mmoi = 45 minutes.
Determine the following:
1. If 25 maintenance tasks were demonstrated, what is the maxi
mum number of tasks that should exceed 30 minutes in duration
for an accept decision?
2. Same as Case 1, but for a reject decision.
TEST METHOD 9 191
where
tc{ = corrective maintenance times,
and
nc = number of corrective maintenance tasks
demonstrated.
"F
W Vp - 4t^>
where
/c = number of expected corrective maintenance
tasks occurring during a representative
operating time T,
fp = number of expected preventive maintenance tasks
tasks occurring during the same
operating time T,
and
fc = XCT and fp = XpT.
Also the following may be used:
E td + E tp,-
1=1 >=1
*c/p =
nc + np
where the td and tpj are determined for the same represen
tative time.
(d) tMmaic = antiloge K + «(*Mm„e)*«i] i
where
Otherwise reject.
194 MAINTAINABILITY ENGINEERING SPECIFICATIONS
Otherwise reject.
Otherwise reject.
EXAMPLE 8-14
For a special, maintainabilitywise vital equipment the following are
specified:
Tc = 5.250 hr,
7* = 1.500 hr,
/? = 0.05 for tc and tp,
t"c/p = 3.000 hr,
and
rMmalc = 12-50 hr' for Mmaxc = 95%.
In a series of maintainability demonstration tests the following are
determined:
nc = 46, tc = 3.595 hr, and a^c = 5.641 hr,
Tip = 60, tp = 1.356 hr, and c^p = 0.375 hr,
tc/p = 2.350 hr, and a^p = 3.135 hr, with fc = 20 and
U = 40,
and
io.95 = 12-08 hr, for MmaXc = 95%.
Determine if the specified requirements have been met by these
maintainability demonstration results.
TEST METHOD 9 195
1. Test for tc
Since
t* = 5.250 hr > 4.963 hr,
we accept the equipment as having met the mean corrective main
tenance time requirement.
2. Test for tp.
— SL 0.375
tP + Z(3-^= = 1-356 + 1.645 —= = 1.436 hr.
v/60
Since
t* = 1.500 hr > 1.436 hr,
we accept the equipment as having met the mean preventive
maintenance time requirement.
3. Test for tc/p.
1/2
nP(fc $tc)2 + nc(fp atp)2'
tc/p + *fi
ncnp(fc + fp)2
"60(20 x 5.641)2 + 46(40 x 0.375)2'"I 1/2
= 2.350 + 1.645
46 x 60(20 + 40)2
= 2.809 hr.
Since
rc/p = 3.000 >> 2.809,
we accept the equipment as having met the requirement for the
mean maintenance time of the combined corrective and preven
tive maintenance actions; i.e., the mean of all maintenance ac
tions.
196 MAINTAINABILITY ENGINEERING SPECIFICATIONS
Confidence level
75% 90%
Acceptance level
22 20
is less than or equal to that shown in Table 8.7 and/or Table 8.8,
corresponding to each index for the specified confidence level.
EXAMPLE 8-15
The following are specified for a specific equipment:
1. Median value of the corrective maintenance times =
ic = 2.75 hours.
zl fpi • XPi
tp = //p(actual) = ,"~1
h
E/P.
where
fPi = frequency of occurrence of the ith task in the
referenced period,
k = number of different preventive maintenance tasks
performed,
y^ /p, = total number of preventive maintenance tasks
200 MAINTAINABILITY ENGINEERING SPECIFICATIONS
performed,
XPi = midpoint of the time intervals in which the
different maintenance tasks were performed,
/ip( actual) = actual mean preventive maintenance time
calculated using this formula = tp.
Accept if 7p(/ip) required > tp(np) actual.
Reject iitp(fip) required < tp(nP) actual.
TABLE 8.10- Calculations for Table 8.9 data to obtain /xp (ac
tual).
Frequency of
Time preventive Class
interval, maintenance tasks, midpoint,
Class minutes f,i Xpi /, >i " Xpi
1 0 - under 10 1 5 5
2 10 - under 20 4 15 60
3 20 - under 30 5 25 125
4 30 - under 40 7 35 245
5 40 - under 50 8 45 360
6 50 - under 60 3 55 165
t=6 t=6
it = 6 N = £ hi = 28 ■ Xpi = 960
t=i
202 MAINTAINABILITY ENGINEERING SPECIFICATIONS
^p(actual) = x—^ ,
where
/Pi = frequency of occurrence of the ith task in the
referenced period,
k = number of different preventive maintenance
tasks performed,
)] fPt = total number of preventive maintenance tasks
performed,
A'Pl = midpoint of the time interval in which the dif
ferent maintenance tasks were performed,
/ip(actual) = actual mean preventive maintenance time
calculated using this formula.
The 27th ranked value is 55 minutes and the 28th ranked value
is 57 minutes. Consequently, the duration of the 27.55th obser
vation, or of the preventive maintenance task time, is
55 + (27.55 - 27.00)(57 - 55) = 56.10 min.
Since
*A/maxp(re(luired) = 58 min. > tMmaz
maxp factual)
\ = 56.10 min
PROBLEMS
where
t = mean time to restore the equipment,
and
at = standard deviation of the times to restore.
w here
t' = \oget.
Determine the additional maintenance tasks that have to be
demonstrated.
3. If at the conclusion of the required maintenance tasks the
following updated results are obtained:
t = 2.9 loge minutes,
and
ati = 1.19 loge minutes,
Determine if the equipment has met the desired mainte
nance time requirement of 60 minutes.
4. Same as in Case 3 but for the maximum allowable mainte
nance time of 75 minutes.
or
T = 10,000 hours,
the average total man-hours that will be required for all scheduled
preventive maintenance tasks will be
PS = 120 man-hours.
If the producer's risk, a is specified as 5%, has the radar sys
tem met its MHR requirement of 0.075 man-hours per operating
hour?
-13. It is specified that the maintainability of the equipment be demon
strated on a dual requirement basis: (1) A mean time to restore
of 30 minutes. (2) A time for a maintainability of 95% of 45
minutes.
Do the following:
t' = loge*.
Determine the additional maintenance tasks that have to be
demonstrated.
3. At the conclusion of the required maintenance tasks the
following updated results are obtained:
t = 3.1 log,, minutes,
and
oti = 1.0 logg minutes.
Determine if the equipment has met the desired mainte
nance time requirement of 50 minutes.
and
«*-,« - 15 hr, for Mmaxc = 95%.
Xa - Mo
= za, (8A.5)
do/y/n
Xa = Mo + za{d0/yfii). (8i4.6)
Similarly, Eq. (8/1.2) yields
X. = mi - *fi{dxl^/n). (8A.7)
Equating Eqs. (8/1.6) and (8A.7) yields
and
a^ = ^e*2 - l)1'2, (SA.ll)
where fi and cr denote the mean and standard deviation of log X.
For E{X) = /x0, from Eq. (8A.11)
Test Plan 5
For Test Plan B
d\ = do = d (8A.15)
where d is the prior estimate of the standard deviation of the mainte
nance time. Substitution of d for do and di in Eq. (8A.9) yields
. 2
za + zp '
n= (8A.16)
mzpL
d
_.l
218 MAINTAINABILITY ENGINEERING SPECIFICATIONS
and
P(\ogeXp<\ogeXpa\H1) = 0, (85.4)
also
(85.5)
and
n- 1
log£ Xp = Y + zp S (85.10)
where
Y = ±±logeX„
S> = -±-£(\ogeX>-Y)\
n — 1 .—;
APPENDICES 219
log«A'
~ =Var F + z,
V n
• Var(S). (85.16)
l°ge A'p n
Also if n is large, then
„2
Var(5) S< — [1, pp. 137-139], (85.17)
and
n- 1
= 1.0. (85.18)
log**? n p 2n (85.19)
_l.
220 MAINTAINABILITY ENGINEERING SPECIFICATIONS
and
X' = EQDg.Xp\Hi) - zfi alogJj>, (85.21)
Since
5(loge XP\H0) = loge T0, (85.22)
and
E{\ogeXp\n1)=\ogeT1, (85.23)
from Eq. (85.19)
(\ n 1 \1/2
a. ? = <r - + zl — . (85.24)
X* = ]0B.ri-*»(i+jjJL) . (85.26)
Equating the right hand side of Eqs. (85.25) and (85.26) yields
»=fl
\ + ^V2(i
2) ^g tV-
Viogeri-ioge7oy v(8^.28);
REFERENCE
1. Duncan, Acheson J., Quality Control and Industrial Statistics, Richard
D. Irwin, Inc., 992 pp., 1974.
APPENDIX 8C - TEST METHOD 3
If X\, X2, . . . , Xn is a random sample indicating a series of main
tenance actions,
0, if the maintenance action is completed before
A',- = ^ the specified maintenance time, T",
1, otherwise,
APPENDICES 221
i= l,...,n, (8C.1)
and
P(Xi = 0) = l-p, P(Xi = l) = p. (8C.2)
and
and
<:-" Pi
V" Pi (J -Pi)
p I S-Y, -npi < c-npi
V\/npi (1-Pi) ~ \/npi (1-Pi)/ y V27F
— oo
ft, (8C.9)
that is
c — n p0
= zn (8C.10)
s/n pa (1 -p0)
and
c — n pi
= -Z0, (8C.11)
s/npi (1-pi)
Rearranging Eqs. (8C.10) and (8C.11) yield
ZaVPO Qo + zpVPl Q\
n = (8C.16)
Pi -Po
Equation (8C.16) gives the sample size needed. Substitution of Eq.
(8C.16) into Eq. (8C.11) and rearranging Eq. (8C.11) yields the ac
ceptance critical value, c, as
Z0 PoVPlQl + Za Ply/poQo
c= n (8C.17)
Za\/PoQo + Zfiy/piQi
APPENDICES 223
(8C.18)
r=0
and
(8C.19)
r=0
Solving Eqs. (8C.18) and (8C.19) yields the sample size n and the
acceptance critical value c.
APPENDIX 8D - TEST METHOD 4
If the underlying distribution of the corrective maintenance task
time, X, is lognormal, and X\yX^ •••,JST„e is a random sample from
X, then Y = logA is normally distributed and y\ — logA"i,$/2 —
log Xi , • • • , ync = log Xnc is a random sample from Y .
Define
(nc \ */"c
II ^' )
and
ERT = specified equipment repair time.
Then
and
5= -lite-!/)2. (80.2)
1 1=1
Assume
}' = log10 A ~iV(/i, a2), (80.3)
then
iV(0,l), (80.4)
r/v/n
i_
224 MAINTAINABILITY ENGINEERING SPECIFICATIONS
^--X'K-I), (80.5)
a2
and
Y-fi , ncS2
—7—7= and —=— are independent.
Consequently,
Y-M — _
/ atg 5
1)
is Student's t distributed.
Here
H = log10 ERT, (8D.7)
therefore
log10MTTRG-\ogwERT
T=-^ 1 212 L_V/^TT „ t(nc _ !). (82).8)
or
log10 MTTRG < log10 ERT + 0.397(5). (80.12)
APPENDICES 225
then
and
P(X<Xa\n = M!) = P, (825.12)
where
1 n
n = sample size
and
Xa = acceptance critical value.
Considering the alternative forms of Eqs. (8JE.11) and (822.12):
and
Pl±Z*L<*!LlM1\li = Ml)=0,
(8jB.14)
c/y/n a/y/n
__L_
228 MAINTAINABILITY ENGINEERING SPECIFICATIONS
and
X. = Mi - ^4=- {BE. 20)
/Mi-MpV
(8F.3)
*<*'7s
and
1
MTBF L
i.+2 TV .MTBF VM+ n J V»» VMT^fJ. '
(8F.4)
Furthermore assume that /x^j is the required man-hour rate. The pro
ducer's risk is a, and the acceptance critical value is fj.*R, then
yields
M MTBF J
xc + ¥ -PR
MTBF
MTBF-fi.
<*Mw PR
MTBF^
= 1-0.
(8F.6)
Since
IvTTFP *c + ^ -Mjj
iV(0,l), (8F.7)
MTBF-J^.
Vr ~ Vr _ _ (8F.8)
MTBFs/Z
230 MAINTAINABILITY ENGINEERING SPECIFICATIONS
P(X = 0) = 1 - p, (8G.2)
then
Nn = y. X<
is the total number of items which failed to pass the test, where A',
has a value of either 0 or 1.
The test hypothesis is
H0 : P = P0,
H, :P = P1.
APPENDICES 231
(8G.8)
logef?
rf1 = (8G.9)
^aisar
loge^4
d2 = log P^-Po)' (8G.10)
1-/?
A= (8G.11)
Q
and
5 = 0 (8G.12)
1-Q
1 - FOR PIAN At
Producer's risk: a = 0.06 at k < 0.22. Consumer's risk: /3 = 0.06
at k > 0.39.
Here
A
dl ~2'75
= ~0M= *«
~3-35'
and
d2 = %£ = 3.35.
0.82
The acceptance test criteria are the following:
Accept if
A^„ < 0.305 n - 3.35. (8G.13)
Reject if
Nn > 0.305 n + 3.35. (8G.14)
APPENDICES 233
Continue test if
0.305 n - 3.35 < n < 0.305 n + 3.35. (8G.15)
2 - FOR PIAN Bi AND PLAN B2
For Plan Bx
a = 0.10,Po = 0.02,
p = 0.10^ = 0.10,
therefore
loge A = logee ^
0.10 = 2.20,
0.10
loge B = lo6 = -2.20,
0.90
log 1-0.10
lu6e 1-002 0.085
c = = 0.050,
W 0.02(1-0.10)
lu°e °-10(1-0-02) 1.69
-2.20
di = T-=r = -1.30,
1.69
and
2.20
= 1.30.
1.69
The acceptance test criteria are the following:
Accept if
Nn < 0.050n- 1.30. (8G.16)
Reject if
Nn > 0.050n + 1.30. (8G.17)
Continue test if
0.050n - 1.30 < n < 0.050n+ 1.30. (8G.18)
For Plan B2
a = 0.10,Po = 0.01,
/? = 0.10, Pi = 0.05,
therefore
loge A = 2.20,
logeP = -2.20,
log
IU6e l^o°i
i_o
05 0.041
i -0.01
0.05(1-0
= 0.025,
10Se 0.0l(l-0
1.651
05
-2.20
dy = = -1.33,
1.651
234 MAINTAINABILITY ENGINEERING SPECIFICATIONS
and
Accept if
Nn < 0.025n- 1.33. (8G.19)
Reject if
Nn> 0.025n + 1.33. (8G.20)
Continue test if
0.025n - 1.33 < n < 0.025n + 1.33. (86.21)
A = if>\. (8H.1)
If n is large
where
Xa = acceptance critical value.
Subtracting fi, from both sides of the inequality in Eq. (8G.10) and
dividing both sides by dfy/n yields
(8HA)
d= (8H.7)
n — 1 f~7
\ 1=1
Accept if
Reject if
Reject if
j
/ic(specified) < Xc + -4=? (8tf.l2)
Reject if
and
ii _ «.
^
Accept if
d2
fiEc + ipm d2 '
n bs.
/Vc(specified) > Ap/c + zp*
^ (/c + fpm) 2 \ r* ^c n.
"pm
(8J5T.20)
Reject if
d2 d2 '
/ip/c(specified) < Xp/c + zp,
Mic+fr^yy'n 'P">,
'pm
(8J5T.21)
Test for tMn
In this test it is assumed that the underlying distribution of the
corrective maintenance task time, A'c, is lognormal, and the pdf of Xc
is
then
loge '««, " A<
(8J/.26)
or
*AW = antiloge (/x + zaa) = e^+z° »>. (8/7.28)
Equation (8/7.28) says that if Xc has a distribution as defined by
Eq. (8/7.22), then its (1 - a)th percentile is equal to antilog (fi + zaa).
Estimating \i and a by
1
M = -£logeX„ (8/7.29)
and
(t^eXi A
antiloge i=l
+ Zo ^TE(^x,.-lEiogcxl.)
V
(877.31)
The test hypothesis is
tMmai <<Mmoi (specified),
and the acceptance criteria are the following:
Accept if
*A/max(sPecified) ^
/ n
Y ^ge A',
E
1 " /
antiloge 1= 1
n
+ Zq\ ^1 ^ V°ge A'! " n1 £
"
l0ge *
(877.32)
APPENDICES 239
Reject if
*Mmoi(sPecified) <
t'=i
antiloge
n \ t=i \ t=i
\
(877.33)
f 1, if U > ts,
l' \0, if *,-<*„
If the hypothesis
77q : tp = t3, (87.2)
is true, where tp is the percentile value of the maintenance distribution,
then P(Xi = 1) = 0.50 for the test for the median if the tp stands for
the 50th percentile, or P(X{ = 1) = 0.05 for the test for the 95th
percentile if the tp stands for the 95th percentile.
So, the test for the median and the 95th percentile is changed to
test the hypothesis
Ho : P = Po, (87.3)
where the value of Po is equal to 0.50 for the test for the median, or is
equal to 0.05 for the test for the 95th percentile.
If hypothesis (87.3) is true considering
r-E*. i=\
(87.4)
240 MAINTAINABILITY ENGINEERING SPECIFICATIONS
£Xi-n P0
i=i
JV(0,1). (87.7)
y/tl P0(l - P0)
Then, from Eq. (87.5),
Ci - n P0
= -^0.25, (87.8)
y/n P0(l - 7>o)
and from Eq. (87.6)
C2 - n P0
= -*0.10. (87.9)
v/n P0(l - 7b)
Substitution of the values of 20.25 an^ ■Zo.io into Eq. (87.8) and Eq.
(87.9) respectively, yields
Cx = n P0 - 0.67^/nP0(l-Po), (87.10)
and
C2 = n 7b - 1.28^/n P0(l - P0). (87.11)
Substitution of n = 50 and P0 = 0.50 into Eqs. (87.10) and (87.11)
yields
CSS 22.63, (87.12)
and
C2 = 20.47. (87.13)
But the acceptance level should be the next smaller integer of C\ and
C2, hence the following table:
APPENDICES 241
Confidence level
75% 90%
Acceptance level
22 20
-I
(87.14)
r=0
Confidence level
75% 90%
Acceptance level
1 0
Chapter 9
PREVENTIVE MAINTENANCE
AND ITS QUANTIFIED
ADVANTAGES
243
244 PREVENTIVE MAINTENANCE
3. r*i Tn -*KV T.
"i
0 M-T, t2+t3+2 Tr
Failures reduced
NF(t1) = NA1 = I
J
0
N(t)A.(t)dL
246 PREVENTIVE MAINTENANCE
ea
'3
pc-
Age, T, hr
or
Recalling that
if we take
then,
■ r=vs-J3***- (9-4)
or
Mr = ggftg. (9.6)
0(7,0
The equations for Ajv(0 an<^ ATrp apply for any component, or
equipment, as long as the R(TP) and consequently R(t) can be formu
lated. It applies for example to equipment with parallel or standby
redundancy when preventive maintenance may be exercised every Tp
hours without aborting the mission.
EXAMPLE 9-1
Consider an equipment consisting of two parallel, constant-failure-
rate units. Find (1) its Rtp and (2) its Mjp.
AGE REPLACEMENT POLICY 249
With maintenance
RTp(t) = [R(Tp)Y R(t),
or
RTM) = (2e-XT>-r-2XT>V
e " " -*Y (2 e"A T - e~A
-2 A T),
T\ (9.7)
where
t = jTp + r, i = 0, 1, 2, - - • , andO<T<Tp.
Assume each unit has a constant failure rate of A = 0.01 fr/hr,
or a mean-life of m = 100 hr.
For Tp -¥ oo, or with no preventive maintenance, j = 0,
RTp^oo(t) = R(t) = 2 e -0.01 t _ -0.02 t
(9.8)
and 0 < t < oo.
1.0 •
0.9 -
V-,.
/- Tp=10hr
0.8-
0.7-
tf 0.6 ■
reliability,
Mission
HOiI
*
o
w .~-Tp = 50hr
0.2"
i \ \ N^-^^Tp =
i \
1 >^s
0.1- Tp-~»^ ■ Tp= 150 hr
- T . hi
' 1 11 1 H 1 1 1
50 100 150 200 250 300 350 400
Mission time, t, hr
or
150 + 50 e-°-02 T? - 200 e"001 T"
Tp~ 1-2 e-°-01 tp + e-°02 tp
For
°° ; -Wjp = 150 hr = 1.5 m,
Tp = 150 ; MTp = 179 hr,
Tp = 10 ; MTp = 1,097 hr,
and for
Tp = 0 ; MT = oo.
These Mtp values for various Tp periods have been plotted in Fig.
9.4.
252 PREVENTIVE MAINTENANCE
T ,
p T
p
hours hours
10 1097
50 504
100 208
150 179
150
200 - -
150 hours ->
p*
150
or
RrAt) = e-^T^\ (9.13')
254 PREVENTIVE MAINTENANCE
therefore,
RtM) = e"A ' = R(t)\ (9.14)
In other words the reliability of the unit with preventive mainte
nance every Tp hours is not any different than that with no preventive
maintenance, if the unit has a constant failure rate!
If the single unit has an increasing failure rate there will be an
improvement in its reliability with preventive maintenance, because
then,
RtM) = [R(TP)Y R(r) > R(jTp + r). (9.15)
age never is greater than Tp age at end of mission.
For a Weibullian unit, with /? > 1, for example, its reliability for a
mission of t duration with preventive maintenance every Tp hours is
given by
R(t) = e V
(i aasaV
■ ) . (9.17)
It may be seen that
RTp(t)>R(t).
When /3 > 1 the Weibullian unit exhibits an increasing failure rate
characteristic with increasing operating time, hence the benefit of pre
ventive replacement every Tp period of operation.
Let us find Mtp for a single exponential unit. Then, from Eq. (9.5),
!lvR{r)dT Tp m-\
/( dT
'"'* - l-R(Tp) 1-- e-^ *»
I(l-e-^p)
1 _ e-Arp
MTp = -r = m! (9.18)
SINGLE EXPONENTIAL UNIT CASE 255
1
At„ = (9.19)
MT|)'
EXAMPLE 9-2
A bearing is operating in an equipment. It has a Weibull times-to-
failure distribution with the following parameters:
/? = 2.0, r) = 2, 000 hr, and 7 = 0 hr.
1. Find the reliability of this bearing with no preventive mainte
nance for 10, 100, 500, or 1,000 hr of operation.
2. Find the reliability of this bearing with preventive maintenance,
assuming that the bearing is replaced preventively every 10, 100,
or 500 hr.
3. Find the mean life of this bearing with no preventive mainte
nance.
256 PREVENTIVE MAINTENANCE
#(T) = e~^335o> .
For T = 10 hr
/ jo \"
R(T = 10 hr) = e"^,oo0; =0.999975,
for T = 100 hr
for T = 500 hr
/_500_\20
J?(T = 500 hr) = e"^ *™) =0.939413,
and for T= 1,000 hr
I 1,000 \20
#(T =1,000 hr) = e~Uoo0; - 0.7788007.
and
•(*)'
RtM = e w .
258 PREVENTIVE MAINTENANCE
For Tp = 10 hr
• I 10 \20 / r \^°
■Rrp=io(0 = e u,00c" e ^•°00'' .
These -Rrp=io(0 values are given in Table 9.1 for 0 < r < 10 and
j = 0, 1,2 and 3, and are plotted in Fig. 9.6.
For Tp = 100 hr
_, /.lOO,^20 _(_r_\i0
RTp=ioo(t) = e K2fi0° ' e UoooJ .
These i?Tp=ioo(0 values are given in Table 9.2 for 0 < r < 100
and j = 0, 1,2, and 3, and are plotted in Fig. 9.7.
For Tp = 500 hr
■ / 500 \20 I r \i0
RTp=50o(t) = t 3 ^2.000 ; e UoooJ .
These RTp=50o(t) values are given in Table 9.3 for 0 < r < 500
and j = 0, 1,2, and 3, and are plotted in Fig. 9.8.
These #rp(<) values for Tp = 10, 100 and 500 hr are shown plot
ted together in Fig. 9.9. A study of Tables 9.1, 9.2 and 9.3, and
of Figs. 9.6 through 9.9 indicates the substantial improvement in
the mission reliability of these bearings with preventive mainte
nance, or replacement, every Tp = 10, 100 or 500 hr. The smaller
the Tp the greater the reliability improvement.
Table 9.4 gives the reliability of these bearings with no preven
tive replacement in Column 4, with preventive replacement every
Tp = 100 hr in Column 5, and using the approximate Eq. (9.21)
in Column 6. A comparison of the results in Columns 4 and 5
brings out the great improvement of bearing reliability with pre
ventive replacement. A comparison of the results in Columns 5
and 6 brings out the fact that after about the third preventive re
placement of these bearings their reliability is well approximated
by Eq. (9.21) with at least a three-decimal-place accuracy. Fig
ure 9.10 illustrates the difference in the bearing's reliability as
calculated from Eqs. (9.21) and (9.16). It may be seen that
the two equations give essentially the same reliability value, with
five-decimal-place accuracy, at mission times that are a multiple
of rp. The values deviate from each other at other mission times
with a maximum deviation of 0.000616 for t > 3 Tp.
SINGLE EXPONENTIAL UNIT CASE 259
#TP=lo(0
T i=o i=i i = 2 J = 3
0 1,00000000 0.99997500 0.99995000 0.99992500
1 0.99999998 0.99997498 0.99994998 0.99992498
3 0.99999775 0.99997275 0.99994775 0.99992275
5 0.99999375 0.99996875 0.99994375 0.99991875
9 0.99997975 0.99995475 0.99992975 0.99990475
10 0.99997500 0.99995000 0.99992500 0.99990000
-Rtp=ioo(0
r j = 0 J = l 3 =1 J = 3
0 1.00000000 0.99750312 0.99501236 0.99252796
15 0.99994378 0.99744692 0.99495653 0.99247225
30 0.99977502 0.99727859 0.99478855 0.99230476
45 0.99949388 0.99699814 0.99450880 0.99202571
60 0.99910040 0.99660565 0.99411729 0.99163518
75 0.99859474 0.99610125 0.99361415 0.99113323
90 0.99797705 0.99548510 0.99299954 0.99052021
100 0.99750312 0.99501236 0.99252796 0.99004983
260 PREVENTIVE MAINTENANCE
c 1.00000
0.99990 -
Fig. 9.6 - Reliability versus mission time plot for the bear
ing in Example 9-2 maintained preventively ev
ery Tp = 10 hr.
SINGLE EXPONENTIAL UNIT CASE 261
1.000
w 0.999 -
a.
fr
ee 0.998 -
0
u
1
m 0.997 ■
c
to 0.996 '
■
>
0.995 *
>a
a 0.994 '
0.993 '
>*
0.992 "
0.991
0.990 -
I —I—
100 200 300 J00
Mission time, t = j T + x, hr
Fig. 9.7 - Reliability versus mission time plot for the bear
ing in Example 9-2 maintained preventively ev
ery Tp = 100 hr.
262 PREVENTIVE MAINTENANCE
Fig. 9.8 - Reliability versus mission time plot for the bear
ing in Example 9-2 maintained preventively ev
ery Tp - 500 hr.
mExample
9-2
aibearing
in
the
for
plot
ntime
mission
RtFig.
e9.9
alivneardb-siulsity
hour*
T-
10■
(hr.
no500
100,
pnrmooeaorivevery
Tp
10, =netatinvedl)y
T
J
hour*
tia*.
I
Ion
Mis*
•t,p
264 PREVENTIVE MAINTENANCE
1.000
R(t)
0.987
Fig. 9.10 - Reliability versus mission time plot for the bear
ing in Example 9-2 maintained preventively ev
ery Tp = 100 hr.
SINGLE EXPONENTIAL UNIT CASE 265
RTp=50o(t)
T j=0 i= i J =2 j = 3
0 1.00000000 0.93941306 0.88249684 0.82902912
50 0.99937520 0.93882605 0.88193861 0.82851102
100 0.99750312 0.93706740 0.88029252 0.82695902
150 0.99439080 0.93414364 0.87754590 0.82437880
250 0.98449644 0.92484875 0.86881417 0.81617610
300 0.97775124 0.91851222 0.86286156 0.81058413
350 0.96983919 0.91107955 0.85587921 0.80402482
400 0.96078944 0.90257809 0.84789370 0.79652231
450 0.95063509 0.89303896 0.83893252 0.78810406
500 0.93941306 0.88249684 0.82902912 0 .77880067
t=7+7?rG+1)-
For this bearing
f = 2,000 rQ + i),
T = 2,000 T(1.5),
f = (2, 000) (0.88623),
or
f = 1, 772.46 hr.
4. The mean life of a unit maintained preventively every Tp hours
is given by
Tp 1 - R(TP) '
[*W*M
%(0=
1.0 0 0 0.9 9375 0.9 7503 0.9 68 0 0.9 5012 0.9 4390 0.9 2528 0.991908 0.9 0 50 0.989431 0.987567 0.982652 0.975309
5
equation,
(0.21).
aEq.
p roximate
e-^)P
R(T)
=
1.000000 0.9 9375 0.9 7503 0.9 4391 0.990050 0.984 96 0.97 751 0.969839 0.960789 0.950635 0.939413 0.884700 0.7 8 01
4
3 T,
hr 0 50 100 50 100 50 100 50 100 50 100 100 100
NO
SINGLE EXPONENTIAL UNIT CASE 267
For Tp = 10 hr it is
^tp=io hr ~
_fv^: dT ( 10 \i
1—e 2>°°°
The numerator may be evaluated by numerical integration, using
Simpson's Rule for example. This method yields
9.99994441
M,rP=io hr x _ 0.99997500'
or
MT =10 hr = 399, 998 hr.
or
^rB=5oo hr = M84 hr.
150,000-
140,000-
Tp,
130,000-
hours
V
hours
120,000-
10 399.998
25 160,004
110,000"
50 80,008
100 40,017
100,000-
300 13,383
90,000- 500 8,084
1,000 4,171
80,000
- 1,773
70,000-
60,000-
50,000
Mean tine between failures when the
bearing is subjected to preventive
40,000- maintenance every Tp hours .
10,000
1,772.46 •
V- -1
100
1
200
1
300
1
400
I
500
1
600
1
700
1
800
1
900
1
1000
/ R(t) dr
Jo
gives the mean time between both preventive replacements and correc
tive failures. The total replacement rate, Aj, is therefore given by
XT = —= = Ace-r
+ Ap.
«p. (9.22)
/o " W) ^
Also
Q(TP) + R(TP)
*T = —„t . . '
So"R(T)dT
since
Q(TP) + R(TP) = 1.
Then,
Q(TP)
^c ~ 7T, (9.24)
So" *00 ^
then,
V ^R{r)dr' (9.25)
where
and
EXAMPLE 9-3
^swo= 10 fr/hr
1 ^Q= 50 fr/hr
kSWE= 100 fr/hr
Jti=o
. e-*SE t e~^SWE (lc) g-AsWO («-'l) ft
e-(*>lE+*7Q-*2E+*SWQ-*SW0) h fa.
or
Lo
RTp(t) = [R(Tp)YR(t),
1
2. The mean time between failures can be obtained from Eq. (9.5),
or,
M £'R(T)dT \
MT> 1 - R(TP) '
272 PREVENTIVE MAINTENANCE
. (l _ e0.00006 Tp\
(9.28)
From Case 2,
R(TP) = £(1,000) = 1 - [1 - £(1,000)]
= 1 - 0.6120199 = 0.3879801,
and
r 1,000
/ R(r) dr = 704.20445,
Jo
CORRECTIVE AND PREVENTIVE REPLACEMENT RATE 273
therefore,
0.3879801
0.000551 rp/hr,
704.20445
or
Xp = 551 rp/106 hr.
Q(TP) __ l-fl(Tp)
Ac~ "A
ti>R(T)dT SoPR(r)dr
0.6120199 = 0.000869
A„ = =r^r?-?z _ nnna„n fr/hr,
. ..
704.20445
or
Ac = 869 fr/106 hr.
where
CL = £e-*»&£. (9.29)
d(C-
rp, or T;-, i.e., by solving ffif = 0 for Tv. This is not easy to do
analytically, hence computer and/or graphical solutions are used. Cr-i
is calculated for various values of Tp, over a range that includes T" or
the minimum value of Cf _7, as shown in Fig. 9.13. A minimum will
exist only if Cp < Cc. If Cv > Cc then the minimum cost policy is that
of no preventive maintenance at all!
It must also be ascertained that the equipment's mission reliability
is at least equal to, if not greater than, the reliability goal allocated to
it with the just found value of T* , or that
If the reliability goal is not met, then Tp should be adjusted to the left,
or to a lower value, to obtain a Ac which satisfies the requirement of
Eq. (9.32). This is usually possible without altering the Cj_j very
significantly, because the Ct-I versus Tp curve usually has a shallow
bottom in the minimum cost region, whereby a change in Tp does not
change the Ct-i value significantly, as may be seen in Fig. 9-13 in the
region of Tp.
0 2T„
2.
2T„
where the x's are failures.
This policy is used when maintenance schedules are based on equip
ment operating time rather than unit, component or part operating
total
the
and
sFig.
Tp,
Rmpcost
between
9.13
cearilhenatvedi-nuoanltcsiehv,iep T*,
Tp,
also
is
of
plus
optimum
The
Cj-i-
mcpoariorernvtcetniatvniecve,
hours
T,
smPacrihenvtednuatlnicve,
shown.
T-MIN.
$
CT,
°ri
1 c 2I« E II U
«
C
O
i* oHo
i°
to s
BLOCK REPLACEMENT POLICY 277
A -±
AP ~ m t (9.33)
Ac = ;1 Tp - f^ [1 - Q(t)) dr
±p (9.35)
tiPR(T)dr Tp TpJ?'R{T)dT
Simplifying Eq. (9.35) yields
IoPQ(r)dr
Ac = (9.36)
TpfpR(r)dr'
But this is the lower bound of Ac because if no fai ures occur before Tv
then it is the same case as the corrective failure rate for Policy I.
Therefore,
Io"Q(r)dr
< Ac7/ < om (9.37)
TpS?'R(T)dT ti*R(r)dT
' v ' " v '
Ac lower bound Ac for Policy
for Policy II I and upper
bound for Pol
icy II
Using these Ap and Ac values the necessary spares can be calculated
as before. Also the preventive maintenance period can be optimized.
The total cost will now be
1 Q(TP)
Ct-ii = CjP X + CCc rT,
, (9.38)
So"R(r)dT
Then, to optimize Tp set
d(CT-n)
= 0,
d(Tp)
278 PREVENTIVE MAINTENANCE
and
and
r3P = 0.15 hr ?
1. The stabilized average corrective failure rate of each unit for Pol
icy II is given by Eq. (9.35), or
c"~ff>R(t)dt
^crr — t;
Jo
using Simpson's Rule with 16 intervals. Then,
/■rP2=440 o 05
/ *W * = fTSnnfi (1-°-76797)+0-75 (387.25) = 387.115,
and
1 1
Xa = 00^11C - 777^ = 0.000310 fr/hr.
387.115 " 440
For Unit No. 3
| W fl(i) * = //■3O0 [0.25 e"000045 ' + 0.75 e^)' '5] dt,
/•Tp,=30o o 25
/ R(t)dt = —^r- (1-0.873716) + 0.75 (286.1135),
where
/ e-tjBo) ' dt = 286.1135,
Jo
using Simpson's Rule with 16 intervals. Then,
/■Tp,=300 Q 95
/ #(t)dt = ——^ (1 -0.873716) + 0.75(286.1135) =284.743.
and
A-,
C3 = 284.743 rrr
300 = 0.000179 fr/hr.
'
BLOCK REPLACEMENT POLICY 281
Jp.
N
E(AC, diR+Xpt ftp)
MTTR = i^—^ ,
E(ACl+APi)
_ _I
282 PREVENTIVE MAINTENANCE
Therefore,
0.013051
MTBF = TV
' '
MTBF = l
(0.000289) + (0.000310) + (0.000179)'
or
MTBF = 1, 285.35 hr.
Then,
R(t = 10 hr) = 0.992250, or 99.2250%.
The reliability for a 10-hr mission after an age of 500 hr, without
any maintenance, from Case 8, is 89.4600%, which compares with
the reliability for the subsystem with corrective plus preventive
maintenance found in Case 5 of 99.2250%. This difference is due
to the increasing failure rate of the subsystem in Case 8 with age,
and preventive maintenance being exercised in Case 5 to reduce
this increase in the failure rate. Hence, the great improvement in
the reliability of this subsystem through preventive maintenance
is an increase from 89.4600 % to 99.2250 %, or approximately a
10-percentage point improvement.
284 PREVENTIVE MAINTENANCE
£» R{(r) dr
>
or
Tpl £*> R,(t) dr
foPlQi(r)dr
T2
for
Range
p2max
p2min
Fig.
for
of
Bathtub
9.14
two
parts.
cu—rves
groups
plmax
plmin
hours
T,
Age,
•a
to (X
286 PREVENTIVE MAINTENANCE
£~RjiT)dT
Qj(TP2)
or
^Tp2^R,(r)dT
" Io"2Q:(r)dr
An estimate of the long-term stabilized reliability may then be obtained
from
The choice of Tpl and Tp2 affects both the reliability and the overall
equipment maintenance cost. The optimum Tv\ and Tp2 should then
be chosen to minimize this cost and yet meet the equipment's target
reliability requirements. The total corrective and preventive mainte
nance cost per hour of operation of the equipment would then be given
by
n-(k+s) ,
CT-II= fc.1
£ (Ch—
m>>
)+£
,=1
Cpi rp
Tpi
+ Cct
" mi{Tp{)
k
+ EiCpJ 7^~ + CCJ —^—}, (9.41)
fr{ JP2 rrij (Tp2)
using Policy II. A similar equation can be written for Policy I also. The
optimum Tpj and T*2 can be obtained now by piecewise optimization,
or by plotting the terms ^2 and J2 separately, versus Ct-ii for various
«' j
values of Tp and separately finding T*j and T*2, as shown in Fig. 9.15.
Usually, further savings would be obtained by making Tp2 an integer
multiple of Tpi. Tpi may then become a minor overhaul and Tp2 a
major overhaul, preventively. Again, it should be checked to see that
the equipment's reliability goal is being met with these Tpl and T*2
preventive maintenance schedules.
If all components in the equipment are reliabilitywise in series, then
N
t'=l
and
For Unit 1
Ct-ii = (10.00)
For Unit 2
Cr-/I = (35.00) -L
ip2
For Unit 3
Ct-ii = (12.00)
rp3
(80.00) (l- (0.25) e"'000045) T" + (0.75) e"^ *| 1
+ £" (0.25) e-(°00045) T + (0.75) e_l*
MOO ' dr
The costs associated with the various preventive replacement pe
riods, Tpi, are shown in Table 9.6 and are plotted in Fig. 9.16
from which it is found that T*x = 132 hr, T*2 = 285 hr and
T;3 = 213 hr.
2. To achieve further savings, T*2 and /or T*3 can be made an m-
teger multiple of T*t . Since
T'p2 _ 285 hr
= 2.1591,
1 pi 132 hr
and
]a __ 213 hr = 1
,6136.
T,pi 132 hr
pfrmpuplus
of aeof
cThe
total
rnFig.
iov9.16
asaeccost
rtvenitcaon-itcveive
Example
9-5.
each
in
unit
for
smcaihnet dnaunlce ,
Tp,
hr
scmPahrienvtdeunaltniecv,e
T,
tO to
REPLACING GROUPS OF PARTS 291
are close to two, further savings are possible if we take 7p"2 and
r;3 to be
Tp2 — 2 Tpj,
and
t;3 = 2t;v
Making T*2 and T*3 integer multiples of T^ will minimize the
number of times the equipment must be down for preventive
maintenance.
3. To check to see whether or not the reliability goal of the sub
system is being met with these minimum-cost preventive main
tenance schedules, calculate the equipment's reliability from
3
REQ(T,t) = l[Ri(Ti,t), (9.42)
EXAMPLE 9-6
Work out Example 7-2 again, as follows:
4. Repeat Case 2 for the case when the T*2 and T*3 are made integer
multiples of T*t. Make sure that the reliability goal of R = 0.980
is still met when the adjustments in the T* are made.
5. Determine the total downtime, plus the corrective, plus the pre
ventive maintenance costs for Policy II for 3,000 cumulative hours
of operation, using the downtimes given in Table 9.7 for each unit
with an overall cost of downtime of $10,000 per hour of subsystem
downtime, and the minimum cost 7£.
6. Repeat Case 5 using the adjusted T^ found in Case 4, the correc
tive and preventive downtimes given in Table 9.7, the preventive
downtimes and costs given in Table 9.8, with the same overall
downtime cost of $10,000 per hour of subsystem downtime, for
3,000 cumulative hours of operation.
7. Discuss comparatively the results found in Cases 5 and 6.
1. Using the data of Example 7-2 the reliability model for each unit
is
and
Rt(T) = 0.25 c-(000045) T + 0.75 e"*** . (9.44)
294 PREVENTIVE MAINTENANCE
and
#3(r;3) = 0.933958,
where
T*pl = 132 hr,
T*2 = 285 hr,
and
r;3 = 213 hr.
or
XEq = 0.00160443 fr/hr.
The subsystem's reliability for 8 hr of continuous operation, using
the minimum cost preventive maintenance schedules, is
Rss = e-x*Q \
RSS{i = 8 hr) = e"0-00160443 <8>,
or
Rss(t = 8 hr) = 0.9872466.
J._
296 PREVENTIVE MAINTENANCE
Since,
Rss{t = 8 hr) > Rqoal,
that is
0.9872466 > 0.980,
then the subsystem is meeting its reliability goal using the up
per bound of the corrective failure rate and the minimum cost
preventive maintenance schedules.
2. The expected number of corrective maintenance actions for 3,000
hr of cumulative operation for Policy II using the upper bound
of the corrective failure rate is given by
Therefore,
Nl- L^giSaJ i
f0piR1(T)dr
N,-±iMSLt
VVC2 — ——; I,
fQ»R2(T)dT
and
i-R3(t;3) t
J*c3 — y
and
Nc3 = (0.000316203) (3,000) = 0.9486.
3. The expected number of preventive maintenance actions that will
be required for each unit for Policy II for 3,000 cumulative hours
of operation, using the optimum T*,, is
t
Npi = Xpi t = —
^, = ^0 = 22.7273,
3,000
NP2 = -^g- = 10.5263,
REPLACING GROUPS OF PARTS 297
and
*.-^ -•"■«•
^EQ = si Kit
t=i
6. Solution 1 - Method 1
The preventive maintenance schedules for Units 2 and 3, T*2 and
Tp3, can be taken as multiples of T*x. Consequently, take
CT = }, Ki (tdd Cd + Cd)
.«=!
x + o Fig.
replaced
Units
afor
Unit
P11
times
times,
m9.17
2
1,
3.
1
graeinvndtets
en-tainvce
3,000
X + 0 i
1
2,772
X + 0
x
X 2.376
+ 0 i
X
XXX + 1
1 1,980
hours,
taand
also
Units
times.
11
2
3
1,
ogether
nd
+ 0
O
mPtimes
arienvtentaincve
XXX
1,584
+ a l
XXX + 0 1,188
+- 1
•
1
■T1-
792
X
\ + o
X
X
x ■f o
396
'l
1)32
264
X + 0
Unit
1 Unit
2 Unit
3
CO
o
REPLACING GROUPS OF PARTS 301
or
Ct- $111,206.06.
Solution 1 - Method 2
Another way of calculating the total corrective and preventive
maintenance, and the downtime cost would be as follows: The
expected number of preventive maintenance actions for each unit
are:
The preventive actions taken for the whole system will be 11.3636
actions for Unit 1 only, and 11.3636 actions for Units 1, 2 and
3 together. The preventive maintenance can be determined in a
tabular form as given in Table 9.9.
Total preventive maintenance cost = $698.87 + $85, 227
= $85,925.87.
The total corrective maintenance cost for the system is given by
operating
mphours
of
the
for
TABLE
Cost
a3,000
9.9
rsystem
ientven-antcieve
Expected downtime,
2.8409 5.6818
6 hr
Dowper
ntime
action,
5 hr 0.25 0.50
mcost,
aintenance
SExample
for
oMethod
2
9-6
1
oluft-io.n Preventive
113.64 585.23 698.87
4 $
Cost
per
action, 10.00 51.50
3 $
Expected of
number
actions 1 .3636 1 .3636
2
the
min
aintenance
iUnits
nvolved
action 1,2,3
1 1
CO
o to
REPLACING GROUPS OF PARTS 303
or
Cc = $28,029.56.
Then,
Total cost = Total preventive maintenance cost +
Total corrective maintenance cost,
or
Total cost = $85,925.87+ $28,029.56= $113,955.43.
Solution 2 - Method 1
Other combinations of T*t, T*2 and T*2 can be used to calculate
the total cost for preventive and corrective maintenance actions;
e.g.,
r;x = 132 hr, T;2 = 3 r;a = 3 (132) = 396 hr,
and
T;3 = 2 2£, = 2 (132) = 264 hr.
The mean corrective maintenance rate for each unit can be cal
culated using
t».
1 - 0.25 e~x> T" - 0.75 e *T?H
A =
rTpi 0.25 e"A« T + 0.75 e -&<
v*><
JO dr
or
XEq = 0.0024166875.
Then
RSS(T = 8 hr) = e-*£<? <8),
JRss(T = 8hr) = e-0-0024166875(8),
or
Rss{T = 8 hr) = 0.98085219 > 0.980.
The total cost can be calculated using the equation
2.772
'-
1
r—I
2,376
792
1,188
1,584
1,980
mParienvtentainvce
hours
times,
T
SaCase
Method
Example
6
2,
1.
9-6,
olndution
1
)
—
1
132
0
396
264
"■'—I
Unit
1 2
Unit 3
Unit
CO
o Cn
306 PREVENTIVE MAINTENANCE
3,000
Unit 3 Np3 = 264 = 11.3636.
The preventive actions taken for the whole system will be 3.7879
actions for Units 1, 2, and 3 together, 3.7879 actions for Units
1 and 2 together, 7.5758 actions for Units 1 and 3 together, and
7.5758 actions for Unit 1 only. The preventive maintenance cost
can be determined in tabular form as given in Table 9.10.
From Table 9.10 the total preventive maintenance cost = $547.36+
$75, 759 = $76, 306.36. The total corrective maintenance cost for
the system can be found by
Cc = z2 ^a {Uci Cd + Cd)
t=i
or
Cc = $43,690.80.
Then
or
Expected downtime,
1.8940 1.5152 2.2727 1.8940
6 hr
Dowper
ntime
action,
5 hr 0.50 0.40 0.30 0.25
macost,
intenance
Sfor
Case
Method
6,
2,
o2.
lution Preventive
195.08 151.52 125.00 75.76 547.36
4 $
Cost
per
action, 51.50 40.00 16.50 10.00
3 $
Expected number
of
actions 3.7879 3.7879 7.5758 7.5758
2
iin
Units
nvolved the
maintenance
CO
308 PREVENTIVE MAINTENANCE
Solution 1 Solution 2
shedules: shedules:
T;x = 132 hr, T;x = 132 hr,
T;2 = Tp3 = 264 hr. r;2 = 396 hr,
T;3 = 264 hr.
Total preventive
maintenance cost
for 3,000 hr,
Method 2 $ 85,925.87 $ 76,306.36
Total corrective
maintenance cost
for 3,000 hr,
Method 2 $ 28,029.56 $ 43,690.80
Total cost for
3,000 hr,
Method 2 $ 113,995.43 $ 119,997.16
Total cost for
Method 1 $ 111,206.06 S 116,707.30
Reliability 98.62% 98.09%
REPLACING GROUPS OF PARTS 309
7. The total cost in Case 5 is higher than the one in Case 6, com
paratively, as follows:
In Case 5 it is $ 143,320.93.
1 SE
*- sw
Xswo= 30 fr/106hr
1 ^SWQ= 50 fr/106hr
?W = 200 fr/106 cycles
PROBLEMS
(4) Plot the MTBFTp found in Case 3 versus Tp for the same
Tp values given in Case 2.
(5) Find the MTBF of this system with no preventive mainte
nance and plot it in the figure of Case 4 for the comparison
purposes. Give all calculations and the results in a neat
table.
(6) Calculate and plot on the figure of Case 2 the reliability of
this system using the expression
Bit) = e'^
using the Mjp value found in Case 4 for Tp = 500 hr. Give
all calculations and results in a neat table.
(7) Comparatively discuss the results of Cases 2 and 6 for Tp =
500 hr.
9-2. Given is the system in Fig. 9.20. Do the following:
(1) Derive the expression for its mission reliability when it is
maintained preventively every Tp hours according to Policy
I.
(2) Calculate and plot the reliability found in Case 1 versus mis
sion time for the following preventive maintenance periods:
Tp = 10 hr, 100 hr, 500 hr, 1,000 hr and oo hr.
Give all calculations and results in a neat table.
(3) Derive the expression for the mean time between failures,
MTBFjp, of this preventively maintained system.
(4) Plot the MTBFtp found in Case 3 versus Tp for the same
Tp values given in Case 2.
(5) Find the MTBF of this system with no preventive mainte
nance and plot it in the figure of Case 4 for the comparison
purposes. Give all calculations and the results in a neat
table.
9-3. Given is a system with three exponential parallel units, each
with a failure rate of 0.002 fr/hr, which is subjected to corrective
maintenance when the system fails and to preventive mainte
nance every 200 hr according to Policy I. Do the following:
(1) Derive the non-maintained mission reliability expression for
this system.
312 PREVENTIVE MAINTENANCE
U-
^0= 10 fr/106hr
XSWQ= 50 fr/106hr
A.SWE= 100 fr/106 cycles
U-
^wo = 10fr/106hr
X.SWQ= 50fr/106hr
XSWE= 100 fr/106 cycles
(2) Derive the MTBF expression for this system when it is not
maintained preventively.
(3) Calculate and plot the reliability of the maintained system
versus the mission time curve for j = 0 and J = 1 only.
(4) Calculate and plot the preventively maintained system's
MTBF versus preventive maintenance schedule for Tv =
200 hr, Tp = 400 hr and T„ = 2,000 hr.
(5) Superimpose on this plot the MTBF of this system when
it is not maintained preventively.
9-5. Solve the bearing problem of Example 9-2 when the parameters
of these bearings' probability density function are the following:
7 = 0.0 hr,
P = 3.0 hr.
and
77 = 3,000 hr.
Plot the reliability function for Tp - 10 hr, Tp = 100 hr, Tp = 500
hr and Tp = 1,000 hr. Then, tabulate your results in the format
given in Example 9-2. Discuss the reliability function's behavior
in detail.
9-6. Solve the bearing problem of Example 9-2 when the parameters
of these bearings' probability density function are the following:
7 = 100.0 hr,
)3 = 2.0 hr.
and
77 = 2,000 hr.
Plot the reliability function for Tp = 10 hr, Tp = 100 hr and
Tp = 500 hr. Then, tabulate your results in the format given
in Example 9-2. Discuss the reliability function's behavior in
detail.
9-7. Given is the system of Problem 7-1, wherein each unit is replaced
preventively. Do the following:
(1) If Unit 1 is replaced preventively every 50 hr, Unit 2 every
200 hr and Unit 3 every 100 hr what is the stabilized average
corrective failure rate of each unit for Policy I?
(2) Same as in Case 1 but for Policy II.
(3) What is the stabilized average preventive replacement rate
of each unit Policy I?
(4) Same as in Case 3 but for Policy II.
(5) What is the system stabilized MTTR (Mean Time to Re
pair) if the mean corrective repair times of the subsystem,
dm, when the ith unit fails, are as follows:
dm = 0.55 hr;
d2R = 0.75 hr;
d3R = 0.33 hr;
PROBLEMS 315
and
F3P = 0.15 hr?
(6) Same as in Case 5 but for Policy II and for both bounds of
Ac.
(7) What is the stabilized MTBF of the subsystem with cor
rective and preventive maintenance for Policy I?
(8) Same as in Case 7 but for Policy II and for the lower bound
of Ac.
(9) What is the stabilized reliability of the subsystem with cor
rective and preventive maintenance for a 10-hour mission
for Policy I?
(10) Same as in Case 9 but for Policy II and for the lower bound
of Ac.
(11) What is the reliability of the subsystem without any main
tenance for a mission of 10 hr?
(12) What is the MTBF of the subsystem without any mainte
nance?
9-9. Given is the subsystem of Problem 7-3 wherein each unit is re
placed preventively. Determine the foDowing:
(1) If Unit 1 is replaced preventively every 225 hr, Unit 2 ev
ery 450 hr, and Unit 3 every 450 hr, what is the stabilized
average corrective failure rate of each unit for Policy I?
(2) Same as in Case 1 but for Policy II.
(3) What is the stabilized average preventive replacement rate
of each unit for Policy I?
(4) Same as in Case 3 but for Policy II.
(5) What is the subsystem stabilized MTTR for Policy I if the
mean corrective repair times of the subsystem, d,/*, when
the ith unit fails, are as follows:
dm = 0.55 hr,
d2R = 0.75 hr,
and
d3R = 0.33 hr,
and if the mean preventive replacement times, t'ip, of the
units are as follows:
F1F = 0.25 hr,
r2p = 0.35 hr,
PROBLEMS 317
and
t'3P = 0.15 hr?
and
t*3P = 0.15 hr?
1 75.00 10.00
2 145.00 35.00
3 80.00 12.00
(3) How would you check to see whether or not the operational
reliability, or the reliability goal, of the subsystem is be
ing met with these minimum-cost preventive maintenance
periods?
9-13. Given is the subsystem of Problem 7-1. Determine the following:
(1) The minimum cost, preventive maintenance period for each
one of the three units using Policy II. The corrective and
preventive maintenance costs are given in Table 9.13. Plot
the total cost of corrective and preventive maintenance for
each unit versus their preventive replacement period, Tp.
(2) Show how it is possible to achieve further savings by making
TP2 and/or Tpz an integer of Tpl.
(3) How would you check to see whether or not the operational
reliability, or the reliability goal, of the subsystem is be
ing met with these minimum-cost preventive maintenance
periods?
9-14. Work out Problem 9-13 again using the lower bound of the cor
rective failure rate for Policy II, and the data given in Tables
9.13 and 9.14 and do the following:
(1) Summarize the results obtained by each unit for each case
of Problem 9-14.
(2) Comparatively discuss the summarized results for each case.
(3) Which top three preventive maintenance schedules would
you recommend and why? List the following:
PROBLEMS 321
9-17. Work out Problem 9-16 again using the upper bound of the
corrective failure rate for Policy II, and the data given in Tables
9.13 and 9.14 and do the following:
(1) If an average, stabilized reliability goal of R = 0.980 is spec
ified for 8 hr of continuous operation, determine whether or
not this subsystem meets this requirement using the upper
bound of the corrective failure rate.
(2) Determine the expected number of corrective (unscheduled)
maintenance actions that will be required for each unit for
Policy II using the upper bound of the corrective failure
rate and the optimum Tp, for 3,000 cumulative hours of
operation.
(3) Determine the expected number of preventive (scheduled)
maintenance actions that will be required for each unit for
Policy II for 3,000 cumulative hours of operation.
(4) Repeat Case 2 for the case when the Tvi and Tp3 are made
integer multiples of Tp\. Make sure that the reliability goal
of R = 0.980 is met when adjustments in Tp, are made.
(5) Determine the total corrective, preventive, and downtime
costs for Policy II using the upper bound of the corrective
failure rate, for 3,000 cumulative hours of operation, using
the data given in Tables 9.13 and 9.14, and overall downtime
cost of $10,000 per hour of subsystem downtime, and using
the optimum Tp,- found in Problem 9.16.
(6) Repeat Case 5 using the adjusted Tp, found in Case 4, the
corrective and preventive maintenance costs and downtimes
given in Table 9.17 and overall downtime cost of $10,000 per
hour of subsystem downtime, for 3,000 cumulative hours of
operation.
(7) Discuss comparatively the results found in Cases 5 and 6.
9-18. Do the following:
(1) Summarize the results obtained by each unit for each case
of Problem 9-17.
(2) Comparatively discuss the summarized results for each case.
(3) Which top three preventive maintenance schedules would
you recommend and why? List the following:
(3.1) Preventive maintenance schedule for each one of the
three units.
(3.2) Number of corrective maintenance actions.
324 PREVENTIVE MAINTENANCE
and
Nun
= 0.90.
Ni
9-20. Work out Problem 9-19 again using the upper bound of the
corrective failure rate for Policy I, and the data given in Tables
9.13 and 9.14 and do the following:
(1) Summarize the results obtained by each unit for each case
of Problem 9-20.
(2) Comparatively discuss the summarized results for each case.
(3) Which top three preventive maintenance schedules would
you recommend and why? List the following:
(3.1) Preventive maintenance schedule for each one of the
three units.
(3.2) Number of corrective maintenance actions.
(3.3) Number of preventive maintenance actions.
PROBLEMS 327
(3.4)
Cost of the corrective maintenance actions.
(3.5)
Cost of the preventive maintenance actions.
(3.6)
Total cost with 7£.
(3.7)
Total cost with the top three Tp,- combinations which
are multiples of each other.
(3.8) Reasons for the top three recommendations.
(4) What other costs should be included in a more sophisticated
analysis and how? Illustrate with hypothetical values.
9-22. Given is the subsystem of Problem 7-2 wherein each unit is
replaced preventively. Assume
= 0.10,
Ni
and
= 0.90.
and
F3P = 0.15 hr?
and
N ■
"" = 0.90.
Ni
f=0.20,
and
= 0.80.
Ni
(3) How would you check to see whether or not the operational
reliability, or the reliability goal, of the subsystem is be
ing met with these minimum-cost preventive maintenance
periods?
9-28. Work out Problem 9-27 again using the upper bound of the
corrective failure rate for Policy II, and the data given in Tables
9.13 and 9.14 and do the following:
6.50 4.50
23.00 15.00
12.00 6.50
REFERENCES
1 . Bazovsky, I., N. R. MacFarlane, R.L. Wunderman, Study of Main
tenance Cost Optimization and Reliability of Shipboard Machin
ery, United Control Corporation, Seattle, Washington, DDC No.
AD 283428, June 1962.
2. Cox, D. R., Renewal Theory, Methuen, London, 135 pp., 1962.
Chapter 10
PERIODIC REPLACEMENT
POLICIES
_ Cp + Cmr E[N(TP)]
Ct-III = 7F, 1 (10.1)
where
335
336 PERIODIC REPLACEMENT POLICY
and
E[N(TP)] = expected number of failures followed
by minimal repair actions in an interval Tp.
The expected number of failures in a Tp interval, per unit, is determined
from
A(r)=|S, ,10.3)
and
R(T) = reliability function of a unit.
Then, the total cost per unit, per unit operating time, is given by
Cp + Cmr £>\{T)dT
Ct-III = ™ • (10-4)
xv
EXAMPLE 10-1
A unit with a Weibull times-to-failure pdf is preventively main
tained under the Ordinary Periodic Replacement Policy. The parame
ters of the Weibull pdf are 7 = 0, /? = 2 and 77 = 30 hr. The cost of the
planned preventive replacement is $5 and the cost of minimal repair is
$30. Find (1) the optimum preventive replacement time, and (2) the
minimal total preventive maintenance cost per unit time of operation.
mmi(£=i)**.-i¥r. (,o.5)
V V 77 )
ORDINARY PERIODIC REPLACEMENT POLICY 337
or
Ztf
f(T) = -7T e<*> (10.6)
IT
\(T) (10.7)
r/2
IfTp x^dT'lfTp2T (T \ 2
iFdT'(V\ (109)
Substitution of Eq. (10.9) into Eq. (10.4), yields the total cost
per unit operating time, per unit, for the OPRP, or
Ie\2
Cp + Cmr(f)
Ct-iii = (10.10)
The minimal total cost per unit time, Cy_//7, and the optimum
replacement time T* can be found by finding the first derivative
of the cost function, given by Eq. (10.10), with respect to the
replacement interval, Tp, setting it equal to zero, and solving it
for Tp. Then,
t; = v\[^- (io.i2)
^mr
338 PERIODIC REPLACEMENT POLICY
T; = 30 ^ = 12.24745 hr.
'cy
(a)
0 T0 + L
cy
(b) -X-
0 T+L
o virtual time
- L to failure
cy
(c) 4- *
T0 + L T
where
T0 = ordering time,
and
R(T0) = probability that a unit will not fail up to time T0.
It is assumed that Ce > CT.
Minimal repairs are made whenever a unit fails. The total expected
number of failures consists of the expected number of failures if the unit
MODIFIED PERIODIC REPLACEMENT POLICY 341
fails in interval (0, T0) and the expected number of failures in interval
(TD, T0 + L) if the unit does not fail in interval (0,To).
Then, the total expected cost of minimal repairs, per cycle, per
unit, is given by
rT0 rT+L
C'T.IV-mr = Cmr Ijf ^ X(x) f(T) dx dT
too rT0+L
+ f(T)dT X(x)dx
J la J In
or
rT0 tT+L
Cx-IV-mr — C„ * / X(x) f{T) dx dT
o Jt
T0+L
Jrio+L.
A(i) dx (10.14)
7a
where
where
C, = salvage value per unit time of operation,
and
f(T) dT = -dR{T).
Using Eqs. (10.13), (10.14), (10.15), and adding the cost of pre
ventive replacement, Cp, the total cost per cycle, per unit, is given
by
C'cy = CP + Ce Q(T0) + CT R(T0)
[ ,T„ ,T+L
+ Cmr jf jT X(x)f(T)dxdT
+ R{T0) f ° \{x) dx
J To
The total expected cost per unit time of operation, per unit, for an
infinite time span can now be determined as
fl(T)=e~('}\ (10.21)
Substituting Eq. (10.22) into Eq. (10.16), the total expected cost
per cycle becomes
- C, r R{T) dT.
JT0+L
The integral
rir-^'-siv-
344 PERIODIC REPLACEMENT POLICY
f(T) dT,
Jo T
V \V/
rT° \2 l
-L'\7Tm+(vm dT,
= /0'^r/(r)<r+(i)|i-*(r.)].
Since Q{T0) = 1 - R{T0) then,
C^ = Cp + Ce-(C;e-CY)i?(T0)
+c~{(^)'-(i)a^+^rT«r>ff
♦«* >[(w-©i}
roo
- C. / fl(T) dT,
or
C'cy = Cp + Ce-(Ce-CT)R(T0)
Substituting Eqs. (10.19) and (10.21) into Eq. (10.23), yields the total
expected cost per cycle, or
lSL)2
C'=Cp
■cy — ^P + Ce-(Ce-Cr)e-{
*V.l «*•--<*!» Aii [ToT"e-02dT
■e K i ' +
T)J Tj V4 Jo
"*i T0+L
e_(^) dT. (10.24)
MODIFIED PERIODIC REPLACEMENT POLICY 345
r~ = z+
Lcy
i: R(T) dT= L +
i: e {"> dT. (10.25)
Tp
■ffl ® ® 1 V V ED »
(k-DTp T0 kTp
and
r?o
E[N1[0tTo]]= I °\(T)dT. (10.28)
"7—&■
«•" "•"
(k-DTD kT„
or
E[N2[To,tp]} = X[Tp-T0-E(t')]. (10.30)
Then, the total cost model for maintenance Policy V, per unit time
of operation , or the total cost rate, assuming an infinite time span, a
constant failure rate for spares, and that at least one failure occurs in
(T0, Tp), is given by
CT-v = Y\Cl+C2J^x^dT
f£R(T)dT
+C3 T -T +1 (10.32)
{* R(T0)
It should be noticed that one in the last term of Eq. (10.32) comes
from the assumption that at least one failure occurs in the interval
(T0,Tp). If the expected number of units failing in that interval is very
small adding one failure assures that at least one failure is accounted
for, conservatively.
348 PERIODIC REPLACEMENT POLICY
EXAMPLE 10-3
Units with a gamma times-to-failure pdf are preventively main
tained under the MPRP, using Policy V. The parameters of the gamma
pdf are (5 = 2 and rj = 200 hr. The scheduled replacement cost is
$15. The minimal repair cost at failure is $70. The spare replace
ment cost at failure is $10 and the constant failure rate of the spares is
X' — 0.05 fr/hr. Find the minimum total preventive and corrective cost
per unit time of operation, Cj_v, the optimum planned replacement
time, T*_v, and the optimum switchover time T*.
SOLUTION TO EXAMPLE 10-3
Theoretically the optimum cost occurs when ^£ and %£?■ are
equated to zero, and T* and T" are found by solving these two equa
tions simultaneously when the other parameters are known. To solve
for the optimum switchover time T*, T* is obtained first using the
OPRP of Policy III, which is a one-parameter policy. Then, substitut
ing Tp* in Eq. (10.32) with the same 13, C\ and C2, T0* will be found as
the value which minimizes further the total cost rate, Cj-v-
To find the T* from the OPRP of Policy III, the gamma times-to-
failure pdf
XW = n rH + ty (10-35)
V {V + P )
The expected number of repairs in a Tp period, per unit in operation,
is given by
Jo V r] + T,
MODIFIED PERIODIC REPLACEMENT POLICY 349
Substituting Eq. (10.36) into Eq (10.1), the total cost per unit time of
operation, per unit, for the OPRP, or Policy III, is given by
or
n
C2
n
T)(V + TP)
-^ + loge(l + ^) (10.38)
The expected residual life, /x', is evaluated from Eq. (10.29) after T0 is
determined, or from
, KRJT)dT
n R(T0)
where
R(T) = eS (l + ^),
350 PERIODIC REPLACEMENT POLICY
(10.40)
e i
Substituting Eq. (10.41) into Eq. (10.31) yields the expected number
of spares, or
^2[to,tp]] = v{tp-T0-^
Tp-T0
+ 2 17 - e" (T, + 2i>)]] (10.42)
Finally, substituting the Eqs. (10.39) and (10.42) into Eq. (10.32) and
dividing by Tp yields the total preventive and corrective maintenance
cost per unit time of operation, or
+ C3 (10.43)
Comparing the minimum cost obtained for OPRP of Policy III with
Policy V, it may be seen that Cf_v = 0.1488 < Cj._//7 = 0.1553.
Therefore, a saving is achieved using Policy V.
MODIFIED PERIODIC REPLACEMENT POLICY 351
* -©- &
(k-DTD
where C\ and Ci are as given in Eq. (10.27), and C$ is the cost rate
of idle time.
To find an expression for E[t], the average residual life at T0, //,
is used again. Then,
E(t) = E{Tp-(T0 + t')\, (10.45)
352 PERIODIC REPLACEMENT POLICY
Finally, substituting Eqs. (10.47) and (10.28) into Eq. (10.44) yields
the following total preventive and corrective maintenance cost, per unit
time of operation, assuming an infinite time span, for Policy VI:
EXAMPLE 10-4
Rework Example 10-3 with a minimum repair cost at failure of
$30. The units are preventively maintained under the MPRP using
Policy VI. The idle time cost is $5 per hour of idle time. Determine
the following:
1. The minimum preventive and corrective maintenance cost per
unit time of operation, CT_VI, the optimum planned replacement
time, T*VI, and the optimum switchover time, T* .
2. The same as in Case 1 for C'z equal to $5, $10, $20 and $30 per
hour of idle time and plot the total preventive and corrective
maintenance cost, Ct-vi, as a function of the switchover time,
T0, for the values of C'z of $5 and $30 per hour of idle time.
To find the minimum total cost, Cf_VI, and the optimum switch
over time, T*, follow the same procedure as in Example 10-3.
The total preventive and corrective maintenance cost per unit
MODIFIED PERIODIC REPLACEMENT POLICY 353
C'z Tp-VI T*
*0 Ct-vi
$/hr hr hr $/hr
PROBLEMS
s
&
•a
g
Switchover time, I, hr
T„ = 304 hr
for C, = S5
minimum total cost per unit time of operation and the ordering
schedule.
REFERENCES
1. Barlow, R. E. and Proschan, F., Mathematical Theory of Reliability,
John Wiley & Sons, Inc., New York, 256 pp., 1965.
2. Kaio, N. and Osaki, S., "Optimal Planned Policies with Minimal Re
pair," Microelectronics k Reliability, Vol. 28, No. 2, pp. 287-293,
1988.
3. Eraclides, S. T., "A Survey and Applications of Alternative Cost Saving
Preventive Maintenance Policies," Master's Research Report submitted
to Dr. Dimitri B. Kececioglu at The University of Arizona, 113 pp.,
1984.
4. Ross, S.M., Applied Probability Models with Optimization Applications,
Holden-Day, San Francisco, 234 pp., 1970.
5. Crk, V., "Optimal Preventive Maintenance Schedule and Spare Provi
sioning Policies," Master's Research Report submitted to Dr. Dimitri
B. Kececioglu at The University of Arizona, 147 pp., 1991.
6. Cox, D. R., Renewal Theory, Methuen, London, 135 pp., 1962.
Chapter 11
MODIFIED BLOCK
REPLACEMENT POLICIES
359
360 MODIFIED BLOCK REPLACEMENT POLICIES
and
/(«) = C[f(T)\ = Laplace transform of f{T). (11.4)
f(T) is the times-to-failure pdf of the unit. Assuming an infinite time
span, or life, of the equipment, the average cost per unit time of oper
ation is given by
EXAMPLE 11-1
A unit has a gamma times-to-failure pdf with (3 = 2 and 77 = 200
hr. The unit is preventively maintained under the ordinary block re
placement policy. The planned replacement cost is $10. The corrective
replacement cost is $50. Find the minimum total preventive cost per
unit time of operation for the OBRP and the optimum replacement
time.
SOLUTION TO EXAMPLE 11-1
With fi = 2, the gamma pdf becomes
'">=(57TIF (u'7)
The Laplace transform of H0{T) is
H0{s) =
5 [1-/00]'
1
•11-^1'
or
H0{s)= l (11.8)
T? sz (77 s + 2)
or
c— £{c'+c-(£+r-'*-i)}- (1112)
The optimum T^ time and the minimal total cost CT_vn can be ob
tained by taking the partial derivative of Cj-vn with respect to Tp,
^f-, setting it equal to zero and solving for T*. Due to the complexity
of the Ct~vii expression, the use of computer optimization will yield
the sought values for T* and CT_VII. The results are
Ct-vii = $0.11874 /hr and T* = 300 hr.
_J_
362 BLOCK REPLACEMENT POLICY
where H0(TP) is the expected number of failures in period (0, Tv). Each
replacement time Tp, is an ordering time as well. The stock of spares
is raised to a level 5, so that the effect of a random demand, which can
cause the shortage of spares, is minimized. If the demand of spares in
the replacement interval Tp is normally distributed with mean, //, then
H = N II0(TP), (11.14)
and variance
a2 = N Var[H0{Tp)}. (11.15)
MODIFIED BLOCK REPLACEMENT POLICY 363
f
j —<
g(x) dx = 1 - a, (11.16)
where
g(x) = distribution of spares demand,
and
1 - a = assurance level of spares provisioning,
or
S-ft'
$ 1-a. (11.17)
fp[H0(Tp)-H(T)]dT
(11.20)
and the holding cost of the average level of spares per unit time of
operation for N units is given by
NCh [Tp[H0(Tp)-H(T)]dT,
(11.21)
Jo
where
Ch = holding cost per unit time of operation per one unit.
364 BLOCK REPLACEMENT POLICY
where
Cb = shortage, or backorder, cost per unit time of operation
per one unit.
If the expected order cost per cycle, Co, is independent of the num
ber of spares ordered; then, the expected ordering cost per unit time
is given by
^- (11-24)
The total cost per unit time is the sum of the costs given by
Eqs. (11.13), (11.21), (11.22), (11.23) and (11.24), or
Cp + Cc H0(TP)
Ctm-viii-\ = N
+^ [Tp[H0(Tp)-H(T)]dT
Jp Jo
+ Ch f (S-x) g(x) dx
Jo
+ Cb (x-S)g{x)dx + -=±,
Js J-v
Cp + Cc H0(TP)
= N
+V^ /"[Ho(Tp)-H(T)}dT
lp Jo
fS rS
+ Ch S g(x) dx -Ch x g(x) dx
Jo Jo
MODIFIED BLOCK REPLACEMENT POLICY 365
+ Cb xg(x)dx-CbS g(x)dx+^,
JS JS in
cP + cc n0(TP)
= N
NCh [Tp[H0(Tp)-H(T)}dT
+
Jo
+ Ch S - Ch S g(x) dx - Ch x g(x) dx
JS Jo
f°° roo
+ Cbj xg(x)dx-CbSl g(x)dx + Co
T '
Cp + Cc H0{TP)
= N
N Ch £P[H0(Tp)-H(T)]dT
+
+ Ch S — Ch S g(x) dx
Js
-Ch U- / x g(x)dx\
TOO »oO (-<
+ Cb xg(x)dx-CbS g(x)dx+^-,
JS JS in
Cp + Cc H0(TP)
= N
N Ch f'Pr
+-~J0 [H0{Tp)-H{T)}dT
Cp + Cc H0{TP)
Ctm-viii-i = N
N Ch f2?,
+ ~^T~ L \HoiTP)-H{T))dT
+ Ch(S-n)- (Ch + Cb)Sa
366 BLOCK REPLACEMENT POLICY
#00 c
+ (Ch + Cb) xg(x)dx + -?. (11.25)
Js lv
NCh rT *
fY}jH0{Tv)-H{T)]dT. (11.31)
kT„
The holding cost of the excess stock and the backorder cost per unit
time are determined by substituting Eq. (11.29) into Eq. (11.22) and
(11.23), or
Cp + Cc H0(TP)
Ctm-viii-2 = N
rTB±_
+irr f"'t^iH o(TP)-H(T))dT
rc ip Jo J=1
+ Ch (Sk - x) g(x) dx
Jo
#oo
+Cb / (x - Sk) g(x) dx
Jst
Co
+ kT„ (11.35)
'Cp + Cc H0(TP)
Ctm-viii-2 = N
368 BLOCK REPLACEMENT POLICY
+tt^ [T'it\jn.(Tr)-n(T)}dT
K 1-p JO .j
+ Ch(Sk-Hk)-(Cb + Ch)Ska
+ (Cb + Ck)J°°xg(x)dx + ^r. (11.36)
EXAMPLE 11-2
100 units with gamma times-to-failure pdf's are put into operation
at time T = 0. They are replaced according to the OBRP on failure and
at a predetermined time interval Tp. The parameters for the gamma
pdf are (3 = 2 and 7/ = 100 hr. The preventive replacement cost is
Cp = $5. The corrective replacement cost is Cc = $80. The shortage
cost per unit time of operation per one unit, is Cb = 20 $/hr per unit.
The holding cost per unit time of operation per one unit is Ch = Si
/hr. The ordering cost is C0 = $100 /order. It is assumed that the
desired assurance level is 95%. (1) If the order for spares is made every
Tp hr, find the minimal total cost per unit time of operation and the
optimum preventive replacement schedule. (2) Repeat Case 1 when
the order for spares is made every 37^ hr.
SOLUTION TO EXAMPLE 11-2
From Example 11-1 the expected number of replacements in a Tp
interval is given by Eq. (11.10), or
*-**(£+k,*-j)- <u-38>
To find the variance of g(x), we have to find the renewal density, h(t),
which is the first derivative of H(t). Then,
*W-yj;(i -«■"*). (1L39)
Substituting H(TP - u) and h(u) into the integral of Eq. (11.28) yields
T 1 fT"\ 1 (T Ax1 "2^ 11 N 2»1 A
I = 2t]Jo
/ [2t}K(Tpp - u)' + -e
4 i 4\ I 1 — e i J du,
MODIFIED BLOCK REPLACEMENT POLICY 369
1 e -*£
•» ti (r,-«) a»
2r]Jo 2 77 4 4 2 7/ 2 7?
_2 i -2 (Tp~M)
e * e '
+ —r-+ : (fat,
or
7= r2p 3 - —re
+— 3 _2Ii Tp e _22i
•> - —— " . (11.40)
8 r?2 4 17 16 16 8 7?
Substituting Eq. (11.40) into Eq. (11.28), yields the variance of the
spares demand distribution, g(x), or
rp , g 2 " 1
*-»&+*?-* 27? 2 t? 4 4
-aS
+2
4 77 + tt-
e 1
8 t?2 16 16
^e"2 . (11.41)
8 7?
The holding cost per unit time of operation for the multi-period
model can be determined using Eqs. (11.31) and (11.37), or
Finally, substituting Eqs. (11.37) and (11.42) into Eq. (11.36), the total
cost model per unit time of operation for N units, and the multi-period
model, is
370 BLOCK REPLACEMENT POLICY
I--*
Cp + Cc(£, + ie
Ctm- VIII-2 = N
J)
| NCh^ T2
kTp i=\ 4 77
D-f(i-l)
•J
T v 4 8' 8J
+ Ch(Sk-»k)-(Cb + Ch)Ska
MO (~>n
x g(X) ax -t- ■ Co
+ (Cb + Ch)Js xg(x)dx + ¥lr. (11.43)
where
£[JVi] = expected number of failures of new components,
E[N2] = expected number of failures of reconditioned
components.
C\ = planned preventive replacement cost,
C2 = failure replacement cost with a new component,
and
C3 = failure replacement cost with a used component.
_ 1
372 BLOCK REPLACEMENT POLICY
—^ © © 1 *? E$3-
(K-l)Tp T0 KTp
and
C~1{ho(s)} = ho(r) in the r time domain.
Another way of defining the ordinary renewal density (ORD), h0(r),
is noting that Eq. (11.47) can be written as
h0(s) = f(s) + h0(s) /(«), (11.48)
and using the fact the Laplace transform of a convolution is the product
of the separate Laplace transforms; then, Eq. (11.48) can be converted
to obtain
ul
VT r'
■—( Au
ib)
o T0 dt1
a
x Denote renewals .
Fig. 11.2- (a) Forward recurrence time, Vt, for Policy IX;
(b) Parameters and variables used in the deriva
tion of the pdf of Vt for Policy IX.
EXAMPLE 11-3
New units have a gamma times-to-failure pdf with 8 = 2 and
rj = 200 hr. The units are preventively maintained under MBRP with
Policy IX. The planned replacement cost is C\ = $10. The failure
replacement cost with a new component is Ci = $50. The failure re
placement cost with a used component is Cj, = $10. The constant
failure rate of the reconditioned units is A' = 0.002 fr/hr. Find the
minimum total preventive and corrective cost per unit time of oper
ation, the optimum planned replacement time, T", and the optimum
switchover time T*.
SOLUTION TO EXAMPLE 11-3
If the new units have a gamma times-to-failure pdf with 8 = 2,
then
T _x
HT): (11.54)
h0(s)
i
(n 5+1)2
" 1 i
U> s+1)2
MODIFIED BLOCK REPLACEMENT POLICY 375
or
1
h0(s) (11.56)
T\ S (7/ 5 + 2)'
Then,
and
/l0(T0-«) = -L[l-e-2iZ^].
(11.58)
2 7?
or
T0 r'
<,. -£.-*{,.-* —+-+1
7/ f? }
U+—+e a
(11.61)
+e~'n-l
Substituting Eq. (11.65) into Eq. (11.50), the mean number of recon
ditioned unit failures in (T0,TP) can be found from
£[JV2M],ATrr0
1 -Iri
H— e Tp2 - Tp ro + 77 (2 Tp - T0 + 2 9)]
(Tp-To)
e (Tp - To)
+1
2 e •» + «""•» - 1
(T, - r0)2
f 2 (T, - T0) + 2 7?
e * —1 + 1. (11.66)
'»
Finally, substituting Eqs. (11.66) and (11.45) into Eq. (11.44), the
total expected maintenance cost per operating period Tp can obtained,
MODIFIED BLOCK REPLACEMENT POLICY 377
or
c™-i{c.+*(.-HSH£-i)+c,{Mr,
- T0 + - e'% [rp2 - Tp Ta + r? (2 T, - Tc + 2 r;)]
(iV-r,,)
2 c" +e
h2(T,-T.) + 2ij
1. Find the T' using the OBRP, as in Example 11-1, and obtain
2. Find the optimum switchover time, and the minimum total pre
ventive and corrective maintenance cost per unit time of opera
tion, and obtain
!
_£=£ Q Q «. £=£__
(K-l)Tp T0 KTp
where
r, = idle time, and {Tp — T0) > r,-.
MODIFIED BLOCK REPLACEMENT POLICY 379
and the total-cost model per unit time of operation, Ct, assuming an
infinite time span, is given by
- I T> W + To)
Jo
EXAMPLE 11-4
Work out the same problem as in Example 11-3, but when the
units are preventively maintained under MBRP with Policy X, and (1)
the idle time cost is $5 per hour, and also when (2) the idle time cost
is $50 per hour.
SOLUTION TO EXAMPLE 11-4
Following the same procedure as in Example 11-3, the total pre
ventive and corrective maintenance cost per unit time of operation, for
an equipment with infinite life time of operation, is
T0 1,.. — 22'">
^a
Ct-x = =r { Ci + C2 (1 - e i )] + C3|j
*{<
- T0 + - e~* [Tp2 - Tp T0 + r, (2 Tp - T0 + 2 V)}
V
1 (TP-T0) ( I
~2e" " {r)l(TP-T0)- + l}
■ £ - m (11.75)
The T* and the Cj_VII using the OBRJP, as was done in the Ex
ample 11-1, are
The optimum switchover time, T*, and the minimum total preven
tive and corrective maintenance cost per unit time of operation are the
following:
(1) C3 = $5 per hour, C?_x = S0.10926 /hr and T0* = 293.75 hr.
(2) C3 = $50 per hour, Ct-x = $0.11874 /hr and T* = 300 hr.
Comparing the minimum cost obtained from the OBRP with that
from Policy X (with the idle time cost $5 per hour), we find that
Cf-vn - 0.11874 > C?_x = 0.10926. So a saving is achieved when
using Policy X. It must be pointed out that there may be no such
savings if the cost of idle time, per hour of idle time, is substantially
large and T0 < Tp. It can be seen that when the idle time cost is $50
per hour the switchover time is equal to the planned replacement time.
That means Policy X can't apply. Therefore the units are actually
maintained under the OBRP.
PROBLEMS
made every Tp hours, find the minimal total cost per unit time
of operation and the optimum preventive replacement schedule.
(2) Repeat Case 1 when the order for spares is made every 4 Tp
hours.
11-4. Work out Problem 11-3 again but for rj = 200, r) = 100 hr and
t] — 50 hr. (1) Find the optimal number of replacement intervals
which minimizes the total cost per unit time of operation. (2) For
each solution of Case 1 determine the number of required spares,
Sk, in stock at the beginning of each interval.
11-5. 100 units with gamma times-to-failure pdf's are put into opera
tion at time T = 0. They are replaced according to the Ordinary
Block Replacement Policy on failure and at a predetermined time
interval Tp. The parameters for the gamma pdf are (3 — 2 and
77 = 100 hr. The preventive replacement cost is Cp = $5. The
corrective replacement cost is Cc = $80. The shortage cost per
unit time of operation per one unit, is Ct = $20 /hr per unit. The
holding cost per unit time of operation per one unit is Ch — $1
/hr. The ordering cost is C0 = $100 /order. It is assumed that
the desired assurance level is 95%. (1) If the order for spares is
made every Tp hours, find the minimal total cost per unit time
of operation and the optimum preventive replacement schedule.
(2) Repeat Case 1 when the order for spares is made every 5 Tp
hours.
11-6. Work out Problem 11-5 again for the desired assurance levels of
90% and 99%. (1) Find the optimal number of replacement in
tervals which minimizes the total cost per unit time of operation.
(2) For each solution of Case 1 determine the number of required
spares, Sk, in stock at the beginning of each interval.
11-7. New units have a gamma times-to-failure pdf with (3 = 2 and 77 =
200 hr. The units are preventively maintained under Modified
Block Replacement Policy-Policy IX. The planned replacement
cost is C*i = $10. The failed unit's replacement cost with a new
component is Ci — $50. The failed unit's replacement cost with
an used component is C3 = $20. The constant failure rate of
the reconditioned units is A' = 0.05 fr/hr. Find the minimum
total preventive and corrective maintenance cost per unit time of
operation, the optimum planned replacement time, T*, and the
optimum switchover time T*.
1 1-8. New units have a gamma times-to-failure pdf with (3 = 2 and 77 =
100 hr. The units are preventively maintained under the Modified
Block Replacement Policy-Policy IX. The planned replacement
REFERENCES 383
11-9. Work out Problem 11-7 again, but now the units are preventively
maintained under the Modified Block Replacement Policy-Policy
X, and the idle time cost is $50 per hour.
11-10. Work out Problem 11-8 again, but the units are preventively
maintained under the Modified Block Replacement Policy- Policy
X, and the idle time cost is $50 per hour.
REFERENCES
1. Cox, D. R., Renewal Theory, Methuen, London, 135 pp., 1962.
2. Eraclides, S. T., "A Survey and Applications of Alternative Cost Saving
Preventive Maintenance Policies," Master's Research Report submitted
to Dr. Dimitri B. Kececioglu at The University of Arizona, 113 pp.,
1984.
3. Acharya, D., Nagabhushanam, G. and Alam, S. S., "Jointly Optimal
Block-replacement and Spares Provisioning Policy," IEEE Transactions
on Reliability, Vol. R-35, No. 4, pp. 447-451, 1986.
4. Crk, V., "Optimal Preventive Maintenance Schedule And Spare Provi
sioning Policies," Master's Research Report submitted to Dr. Dimitri
B. Kececioglu at The University of Arizona, 147 pp., 1991.
Chapter 12
ADDITIONAL MAINTENANCE
POLICIES !
385
386 ADDITIONAL MAINTENANCE POLICES
Opportunities
Residual life
<— after crossing the
— threshold age, Tpxi
Operating time
i
388 ADDITIONAL MAINTENANCE POLICES
where
mean total cost in one renewal cycle
= mean total corrective replacement cost before Tp-xi is reached
+mean total preventive and corrective replacement cost
after Tp_x/ is reached,
or
mean total cost in one renewal cycle
= N Ce + [(1 - u I) Ce + % • J-Cp],
and
mean length of one renewal cycle
= mean time to the preventive replacement age
-f mean time the component remains above the
preventive replacement age,
or
mean length of one renewal cycle
= L + l,
and
Cp = preventive replacement cost,
Cc = corrective replacement cost,
Tp-xi - prescribed threshold age, or preventive replacement
age,
N — mean number of failures before the preventive
replacement age, Tp-xi, is reached,
or
1 - R(Tp.XI)
N = (12.1)
R(TP-xi) '
L — mean time to the preventive replacement age,
or
fo>-x'R(t)dt
I = (12.2)
R{TP-xi) '
/ = mean time the system remains above the preventive
replacement age,
390 ADDITIONAL MAINTENANCE POLICES
or
and
u ■ I — (mean number of preventive replacements above Tp-xi) ,
< l(at most 1).
Then, the mean total replacement cost per unit time of operation,
C(Tp-xi), is given by
C(TP_X/) = [N-Cc + (l-u- l)Cc + u-l- CP}/(L + I). (12.4)
Equation (12.1) can be derived as follows:
The number of trials to reach the specified preventive replace
ment age Tp-xi is a geometric process with the success probability
of R(Tp-xi) and the failure probability of 1 — R(Tp-xi) f°r each trial.
Note that the last trial is always a success trial preceeded by several
failure trials. Each one of these preceeding failure trials consists of a
failure event occurring prior to age Tp followed by replacing the failed
unit by a fresh one. The success trial always ends up with a unit still
surviving at age Tp. The mean number of failures before a specified
age Tp-xi is reached, N, is given by
N = f^k{[l-R(Tp_Xj)]kR(Tp-xi)},
00
= j>[l- XiT,-xi)]k RP^xi),
jt=i
or
Since
1 - R{Tp.Xi) = F(Tp-xi),
then
oo
TV = R(TP.XI) F{Tp-xi) X> [F{Tp_XI)]k-\
k=\
= R(Tp.xi) F(Tp-xi) Jtjj^J -JF(TP-X])]h,
£J dF{Tp-xi)
OPTIONAL REPLACEMENT POLICY 391
A I
= R{TP.XI) F{Tp.xl) dF{Tp_xi) [Y,[F(Tp.XI)}k
F{TP.XI)
= R(TP.XI) F(TP_XI)
dF(Tp.XI) 1 - F(TP,XI)
1-F(TP-XI) + F(TP.XI)
= R(TP.XI) F(TP.XI)
[1 - F{TP.XI)Y '
= R(TP.XI)[1-R(TP.XI)} l
[R(Tp-xiW
or
1 - R(TP.XI)
N =
R(TP.XJ)
Equation (12.1) can also be derived directly as follows:
The mean number of trials to reach the preventive replacement
age Tp_x/ is the mean of the geometric distribution with the success
probability of R(Tp^Xj) for each trial, or
1
R(Tp-XiY
Since the last trial in a geometric process is always a success trial,
then the mean number of failures before the preventive replacement
age Tp-Xi is reached, N, is given by
N = 1 -i,
R{TP-Xi)
or
1 - R(TP.XI)
N
R{TP-xi) '
Equation (12.2) can be derived as follows:
The time for the component to reach the preventive replacement
age Tp-xi, starting at age of zero, is the so-called first passage time.
The mean time to reach the preventive replacement age Tp-Xi, L, is
the sum of the mean time of the first N failure trials and the length of
the success trial (the last trial) which is always Tp-Xj; i.e.,
L = (mean number of failures before the preventive
replacement age Tp-Xi is reached) x (mean
conditional life for each failure trial) + Tp-Xj,
1 - RjT^xi) [Jop-XITf(T)dT
+ TP ■XI,
R{TP-Xi) 1 - R(TP.XI)
..i. _
392 ADDITIONAL MAINTENANCE POLICES
or
fop-XI T f(T) dT
L = + Tp-xi-
R(Tp-xi)
The term for the mean conditional life is actually the mean life of the
truncated time-to-failure distribution or f(T) truncated at Tp-xi- But
/ T f(T) dT = T d[-R(T)]
Jo Jo
Integrating by parts yields
'rT, T d[-R(T)] = -T R(T) P + / R(T) dT,
o ° Jo
or
/ " X' T f(T) dT = -Tv.xi ■ R(Tp-xi) + [ '"*' R(T) dT.
Jo Jo
Therefore,
_ -Tp.xi ■ R(Tp-xi) ± I0p-XI R(T) dT
L = + TtP-xi,
R(Tp-xi)
-Tp-xi ■ R{Tp-xi) + Tp-xi • R(Tp-xi) + Jop-XI R{T) dT
R{Tp-xi)
or
fr-XI R(T) dT
L =
R{Tp.xi) '
Equation (12.3) can be derived as follows:
After the preventive replacement age Tp-xi iS reached, the com
ponent will be replaced either correctively whenever it fails, or pre
ventively once the opportunity occurs, whichever comes first. This
replacement process can be considered as a "reliabilitywise-in-series"
process with the "component failure" and the "opportunity arrival"
as its "components." Then, the process reliability in a mission time t,
after the preventive replacement age Tp-xi is reached, is given by
R(Tp_xi ± t) -u t
R process w- R(Tp-xi)
where e~u ' is used for the residual (conditional) life reliability of the
opportunity, since the exponential distribution is assumed for the time-
to-opportunity arrival.
OPTIONAL REPLACEMENT POLICY 393
Therefore, the mean time the component remains above the pre
ventive replacement age, /, is given by
MO
/ = / -%>rocess(0 dt,
or
-l R(Tp-xi + t)
R(TP-xi)
e~u ' dt,
1 e ■»,
VT(10) \v r
or
TV-
f(T) — I e i .
r?(9!)
- = 0 « = 10 ».
u
2. Determine the effect of different replacement opportunities on
the best preventive replacement age.
!
3. Discuss the results.
__.!___
394 ADDITIONAL MAINTENANCE POLICES
1. Figure 12.2 shows the average cost rate for an Optional Replace
ment Policy expressed as a ratio of the cost of implementing an
Optional Replacement Policy to that with no planned replace
ments at all (pure corrective replacements). Separate lines have
been drawn for various ratios of Cp/Cc- Examination of Fig. 12.2
shows that for some combinations of costs and the preventive re
placement age, the performance of the best optional replacement
policy is considerably cheaper than either the Block policy (which
corresponds to a preventive replacement age of zero), or a policy
with no preventive replacement at all (which corresponds to a
preventive replacement age of infinity).
2. The effect of different frequencies of replacement opportunities
on the best preventive replacement age is shown in Table 12.1.
The top row of the table applies when the replacement opportu
nities are rare compared with the frequency of failures, and the
bottom row when the replacement opportunities are relatively
OPTIONAL REPLACEMENT POLICY 395
frequent. In each row block there are three figures for each cost
ratio. The center figure gives the best value of the preventive
replacement age, and the upper and lower values give the preven
tive replacement ages resulting in costs in excess of the optimum
by 5%. At the bottom of the table for comparison purposes is
the best preventive replacement age for an age replacement pol
icy which is the limiting case of the optional policy with "very
frequent replacement opportunities."
3. The following can be concluded from the results in Table 12.1:
(a) The most important factor in determining the best preven
tive replacement age is the cost ratio, Cp/Cc.
(b) The best preventive replacement age increases as replace
ment opportunities become more frequent, because the in
crease in Tp-xi results in more frequent corrective replace
ments at the age lower than Tp-xi and less frequent preven
tive replacements plus corrective replacements above Tp-xi>
due to more frequent opportunities.
396 ADDITIONAL MAINTENANCE POLICES
where
H(Tto\Tq) = expected number of failures in interval Tro, with
immediate replacement, starting with a
component of age To- H{Tt0\Tq) can be
generated by the following recurrence
relationship:
(12.6)
and
Pxro(y|To) = probability of a component having an age of Y
at the end of an interval of length Tro given
that it had an age To initially. Pjro(y|To) can
be generated from the following recurrence
relationship :
OPTIONAL REPLACEMENT POLICY 397
+PTro.l(y|0)^o^o±ili
0,
« = {«:
Po(Y\T0) =
otherwise.
(12.7)
1. Figure 12.3 shows, as Fig. 12.2, the average cost rate for an Op
tional Replacement Policy expressed as the ratio of the cost of
implementing the optional replacement policy to that with no
planned replacements at all (pure corrective replacements). A
major feature of Fig. 12.3 is the step at a preventive replacement
age ratio of 1.00. This occurs because consecutive replacements
are prevented when the preventive replacement age exceeds the
interval between planned replacements.
Comparing Fig. 12.3 with Fig. 12.2 yields the following conclu
sions:
(a) The costs are lower when the replacement opportunities are
regular and the same preventive replacement age is in force.
(b) The best preventive replacement age for any cost ratio when
the replacement opportunities are regular is lower than that
for the case when replacement opportunities are exponen
tially distributed.
i _
398 ADDITIONAL MAINTENANCE POLICES
Cp/Gc
Ratio C*Ht 0.5 0.3 0.2 0.1 0.05 0.01
0.25 0.90 0.70 0.60 0.50
0.40 0.20 0.10 0.05 0.00 0.00
0.00 0.00 0.00 0.00 0.00 0.00
1.00 0.96 0.58 0.42 0.25 0.21 0.17
0.46 0.25 0.17 0.06 0.02 0.00
0.05 0.03 0.00 0.00 0.00 0.00
l
400 ADDITIONAL MAINTENANCE POLICES
CL = i2e~"spi-^r-- (12-10)
,=0 J-
and
INT(t • u) = integer part of (t ■ u) rounded
to the next lower integer value.
(b) If Tv.xi > I/", let k = INT(Tp-xi ■ «j and
Then,
Tl(t) £ [R(K/u)YR(t), (12.12)
where
j = INT{t/(K/u)},
and
r = < - (ir/U)i.
Then,
7e(0 = [i?(A'-rro)p'.JR(r), (12.14)
where
j = INT(t/K-Tro),
and
T = t-(K-Tro)-j.
402 ADDITIONAL MAINTENANCE POLICES
1). New items (not burned-in yet) are used to replace those that fail
in the test-bed (Stage 1), while the in-service failures (Stage 2) are
replaced by items already operating in the test-bed (Stage 1). The
Two-stage Replacement Policy can be applied to such a system. This
process has been generalized to "multi-stage burn-in" where more test-
beds each with a certain number of items are used. The stage order
number for each test-bed is assigned according to its burn-in length; i.e,
the test-bed with the shortest burn-in length is Stage 1, the test-bed
with second shortest burn-in length is Stage 2, • • •, the test-bed with
the longest burn-in length is the last stage, say Stage n. New items
always go to the vacancies caused by the failures in Stage 1. Service
failures are always replaced by surviving items from Stage n. Failures
at Stage i (1 < i < n) are always replaced by surviving items from
Stage (i - 1).
If the failure rate of the items increases with age, that is, the older
the item becomes, the more likely it is to fail, and Cj\ > Cji because
failures among the first group items incur extra costs by causing dam
age to other parts of the system, then it seems reasonable to suppose
that the following strategy would reduce the overall replacement costs.
Replace all failures in the second group (hereafter called Stage 2) by
items already operating in the first group (Stage 1). Fill all the vacan
cies in Stage 1 by new items, whether the vacancy is caused by failure
or by transfer to Stage 1. It is clear that such a strategy cannot affect
the overall failure rate, or the failure rate of the whole system. It will,
however, decrease the failure rate in Stage 1, where replacement of fail
ures is relatively expensive, at the price of increasing the failure rate
in Stage 2 where replacement is cheaper. If we assume that the cost of
transferring an item from Stage 1 to Stage 2 is small compared with
Cji, such a policy will result in a net saving over simple replacement.
A strategy of this kind will be described as "two-stage replacement."
It may be represented diagrammaticaUy by Fig. 12.4.
404 ADDITIONAL MAINTENANCE POLICES
STAGE 1 Failures
New
components in Stage 1
or
n-l
Cn = [(N/m - ft Pit2)Cfi + £(tf,-i K-U ~ Ni P«.«+i)C/-
t=2
n-l
+JVn_i />„_,,„ Cjn] + Yl Ni P«. •+! C«. »+l
1=1
+(N/m)Cc, (12.15)
where
or
and
Cc = capital cost of a component.
The transfer rate, P{i t+i, depends on the transfer rule and the
life distribution of the components considered, and is determined as
follows:
1. If transfer is made by age; i.e., the oldest component in Stage i
is transferred to Stage (i + 1), then P,-, ,+i can be obtained by
eliminating T, between the following two equations:
+ Nn)/ZN
t=i (12.1/)
Ni Pi, i+, = (5) R(Ti),
where T, is the critical age for a component in Stage i. The
critical age for a component in Stage i is defined as the limiting
age of the components in Stage i when the system approaches its
steady state; i.e, when the system approaches its steady state,
the age of each component in Stage i will be less than T,. Solving
Eq. (12.17) yields
E "i
Pi, i+1 - Ni m
E »i (12.18)
Ti R -l ~7T
V
where R-1( ) is the inverse function of a component's reliability
function.
2. If transfer is made at random, then P\t 2, Pi, t+i(i = 2, • • • , n — 1 )
and Pn_it n are given by
' /0°° R(T)e-^T dT = (Ni/N)m,
< /0°° R(T)e-p>,<+>T dT = Ni/(Ni-i P,-i, ,), (12-19)
/0°° R(T) dT = Nn/(Nn-i Pn_!, „).
EXAMPLE 12-3
Assume TV components have an exponential life distribution with
failure rate A. Find the transfer rate P1( 2 for the Two-stage Replace
ment Policy with N\ and iVj components at Stage 1 and Stage 2,
respectively, if
MULTISTAGE REPLACEMENT POLICY 407
j=i+i _ N2
Pl,2 =
Nxm ~ Nx (1/A)'
or
N2
Pi ,2 = -rr- A,transfers/(component)- (hr).
Jo N
or
1 AT,
A + Pi,2 A N
Then, since N = Nt + N2,
Pi-2 - ~iv7"A - n; a-
EXAMPLE 12-4
There are ten (10) interchangeable, non-screened, line-replaceable
units (LRU's) in a complex electronic system. According to their
failure replacement cost, they can be divided into the following two
groups:
-108 ADDITIONAL MAINTENANCE POLICES
E ^
P - ;=1+1
or
N2
P\2 — t; transfers/(component)- (hr). (12.20)
N\ m
MULTISTAGE REPLACEMENT POLICY 409
Rearranging yields
(P1<2)2 + K Ph2 + G = 0,
where
A' = A, + A2
and
N
G = Aj A2 — [Pi A2 + (1-Pi) Ai].
Ni m
Solving Eq. (12.21) for Ph2 yields
-if + y/K2 -AG
A,2 = (12.22)
and
IO ->
C2 = —— (250 + 200) + 4 (1.5 X 10~3) (600 + 25 - 250),
oi-
C2 = $6.750/hr. (12.26)
where
10
K = 0.01 + 0.000909 - = 0.0084,
4(1,000)
and
10
0 = 0.01 (0.000909)- [0.1 (0.000909) + 0.9 (0.01)],
4(1,000)
or
G = -1.364 x 10"5,
and
10 (250 + 200) + 4 (1.391 X 10_J) (600 + 25 - 250)
1,000
or
C2 = $6.587/hr. (12.27)
/ £(T)[A(T)-(l/m)]dT< 0. (12.29)
JT,
respectively, where
or
Nsp = {N/m)t. (12.33)
The actual number of spares, Nsp, with CL = 1 — a can be obtained
by solving
CL = Yie-Nsp(J^lPl. (12.34)
j=0
OPPORTUNISTIC REPLACEMENT POLICY 413
Since the failure rate of the system is not influenced by the policy,
the system's reliability will be the same as that when no maintenance
policy is implemented [2]. If the N components are reliabilitywise in
series, then
N
K,(t) = H'R.t(t) = [R(t)f. (12.35)
t=i
12.2.7 COMMENTS
Backward direction
I—h
n-1 n-2 n-3 n-4 2 10
Stage or number of periods to operate
Forward direction
I—I—I—I—I 1—I—I
0 A 2A 3A 4A (n-2) A (n-1) A nA=T
:
Accumulated operating time
A = T/n, (12.36)
s. t. SSk-i = tk(dsk,SSk,ek),
dsk e Dk
(12.37)
where
dsk = decision made on the system at stage fc,
e/t = undetermined events which may occur
under ds^ and SSk with their own
feasible range Ek and mass function
Pk(ek), and causes the
next state to be stochastic,
tk{.) = state transition function at Stage fc,
SSk-i — system state at Stage fc — 1 which is
a random variable,
fk(dsk,SSk,ek) = decision cost at Stage fc which is a
function of the system's current state
SSk, decision dsk and the random
events ejt under SSk and dsk,
s. t. = abbreviation of "subject to,"
416 ADDITIONAL MAINTENANCE POLICES
and
Dk = feasible decision sets for the system
at Stage k.
Under the opportunistic replacement policy, the feasible decision sets
at any stage of the system are
replacing Component 1 preventively,
replacing Component 2 preventively,
I C* in all,
Cm in all,
(12.38)
where
Cm = combination value of m choosing i.
The state transitions of any component L, L = 1, 2, • • ■ , m, in the
system with an age of Sfc at Stage k under different decisions are shown
in Fig. 12.6, where k = n - 3 and Sfc = S£_3 = 3. If component L is
replaced preventively at Stage k, then an identical new component will
start its mission immediately at Stage k with an initial age of zero. If
the decision is to "keep it in service," then there will be two possible
outcomes:
OPPORTUNISTIC REPLACEMENT POLICY 417
(•) The Component L with an age of 3. which may be read off the ordinate,
at the corresponding Stage (n-3), as may be read off the abscissa, is
replaced preventively bv an identical new component starting its
mission immidiately at Stage (n-3) with an initial age of zero.
0>) The Component L with an age of 3 at Stage (n-3) is not replaced preventively,
but it fails before the next stage, or Stage (n-4). Then it is replaced
correctively by an identical component starting its mission
at Stage (n-4) with an initial age of zero.
(c) The Component L with an age of 3 at Stage (n-3) is not replaced preventively,
at Stage (n-3), and it survives the next period with an age increment
of 1 , or increases its age to 4 at Stage (n-4).
<f =0
Forward direction
H—I—h H—I—I
2A 3A 4A (n-2) A (n-1) A nA=T
Accumulated operating time
cpm + ck{sl,sl,--,o),
Cpi2 +C*(0,0,Sj»,-", SJ"),
Cpl3 + Cik(0,Sj,0,Sj,.--,S™),
Cl in all,
= MIN <
Cpl3...(m-i)m + ^(0,0, • • • ,0) } C% in aU,
where
Plsl = probability of Component L with present age
Sg surviving the next period,
Cpi 4- PCi = corrective replacement cost of Component
i where i = l,2,---,m,
Cpij + PCij = joint corrective replacement cost of
Components i and j, where i = 1,2,
■ • • , m, j = 1, 2, • • • , m, and i ^ j,
and
Cpi2.(m-i)m + PCi2-(m-\)m = joint corrective replacement
cost of all m components.
By the conditional probability law
PLst = Rl[(S£ + l)A)/RL[(S£)A}. (12.40)
R{T) = exp
-?)' (12.41)
where
P = shape parameter,
and
T] — scale parameter,
420 ADDITIONAL MAINTENANCE POLICES
then
(SL+l) A 1/3'
exp< —
Plsi =
(StL)A
exp ■
v
or
O t C1 C2 CM dsl3 t3lJ3i'J3r"i'J3
/"■ fC1 C2 C"Mj <fcj
Wl-->2> °2' " ' "' °2 /
ds!
^ CkiSioSin-'-yS™).
in unit of hours, or
in unit of A, where
and
Let b; and bu be the lower and upper bounds for the subscripts of array
B. Then, the total number of elements in arrays A and B should be
equal; i.e..
When array A(a\,a2, • ■ -,am) is stored rowwise into array B(b), their
subscripts have the following relationship:
i = l,2,--,m, (12.47)
where
(12.48)
i=f'+i
and
Wm = 1. (12.49)
Conversely when array B(b) is stored rowwise into array A(a,i,a,2, ■ ■ ■ ,am)>
their subscripts have the following relationship:
fc-i
b-bi- Yl Wi(ai - an)
B —> A : ak = J7VT i=l + a/fc,
Wfc
fc = 1,2, , m. (12.50)
where
The feasible decision sets for the subsystem at any stage are
M
PlJ*dl€
$ffWffr*-
Fig. 12.7 - Schematic of a continuous, steelplate, rolling mill
table.
Roll spindle
\ Bevel gear
Bearing
3 W///rWt
Bearing /
"
# 5222 # 97522
R,
Bearing
M
Bearing
#5222 # 97522
Fig. 12.9 - Reliabilitywise series system of two bearings sup
porting a roll.
^•('-^)^{&+te+cc^)•ll,
+(1 - P1S>) ■ (1 - Pis') ■ [C,a + PCn + C,_,(0,0)]}.
(12.52)
A computer program has been developed for the recurrence com
putation of this cost model. The two-parameter Weibull distribution is
used in the program for the bearings' lives. The computer flow diagram
is given in Fig. 12.10.
t
C0(J, K) = 0; Cflo( J. K) = 0; JK = 0
^
Cornpuu C,(0,0)
i
CoopiiuC,(0,K),*r« 1,3 NL(2)
I
|CompuUC,(J,0),J = 1,3 /VL(l)
| J := 0; K :^"o"|
= / + l|^-<^<^>>-
( End )
and
1.33
F(T)\97522= l-exp (—
KSIMJ)
where the unit for T is day.
The length of each period is taken to be A = 30 days, or one month.
Then, according to Eqs. (12.43) and (12.44) the "oldest ages" for these
two bearings are
^nr„.99|5222 + o.5) = /ivrf66-46(los«100)2/3 + 0.5
30
= 6 months,
and
81.6 (loge 100)3/'1
/JVT(To.99|97522 + 0.5) = INT + 0.5
30
= 9 months.
Therefore, from Eq. (12.43) and the above results, the correspond
ing state sets for these two bearings are
S\ - {0,1,2,3,4,5,6} in months,
and
S\ - {0,1,2,3,4,5,6,7,8,9} in months.
_L.
428 ADDITIONAL MAINTENANCE POLICES
Cp" = £l+S2
2 = 362.0+148.5
2 = $255.25,
and
PCl2 . ££1±PC1 = 35,000+70,000 =
and
1. If the two bearings are completely new (with the age of zero), no
replacement will be needed. This corresponds to decision "CR12"
in Table 12.3.
2. If one of them reaches an age older than, or equal to, one (1)
month while the other is less than one (1) month, the replacement
is made only upon the former. This corresponds to decisions
"PR1" and "PR2" in Table 12.3.
3. If both of them share ages which are older than one (1) month,
joint preventive replacement should be conducted on them. This
corresponds to decision "PR12" in Table 12.3.
4. The cost savings in one year by opportunistic replacement against
corrective replacement, which has long been conducted in the
plant, is as high as about 50%, which will be verified next.
The maintenance cost for one subsystem in one year, or for twelve
(12) months, for corrective replacement only is
(12.57)
where
or
MTTF-i, £ 2 months,
and
MTTF2 = mT(l/p2 + l),
- 81.60 r(l/1.33+l),
= 81.60 (0.91906),
= 74.995 days,
or
MTTF-i. £ 2.5 months.
Then, the annual corrective maintenance cost for one subsystem, Ct-c,
from Eq. (12.57) is
12.3.7 CONCLUSIONS
Tp < Tp -
Tp + tp Tp + tp
-4— —► •*— —*■
j = interest rate.
Discounting ^(Tp + tp) back to the start of the first cycle yields
B2~i(Tp + tp) = B2(TP + tp)e~^T"+t^. (12.60)
The total net benefits of the third cycle of operation, discounted
back to their present day value at the start of the third cycle, is given
by
B»(T, + tp) = / "[b(T) e~> T dT] - c(Tp) eH T". (12.61)
Jo
Discounting B3(TP + tp) back to the start of the first cycle yields
J33_i (Tp + tp) = B3(TP + fp)e-,'t2(T"+'"M. (12.62)
Similarly, the total net benefits of the nth cycle of operation, dis
counted back to their present day value at the start of the nth cycle,
is given by
Bn{Tp + tp)= I "[b(T) e-' T dT] - c(Tp) eH T", (12.63)
Jo
which, discounted back to the start of the first cycle, yields
Bn-.i(Tp + tP) = Bn(Tp + tp) e-'(«-D(^+'p). (12.64)
The total net benefits over a long period of time, with preventive
replacements at age Tp, discounted back to their present day value
at the start of the first cycle, is the same as the net benefits over an
infinite number of cycles discounted back to their present day value at
the start of the first cycle; i.e.,
B{Tp) = Bx{Tp + tp)
+B2^(Tp + tp)
+B3^(Tp + tp)
+Bn^(Tp + tp)
or
B(Tp) = B1(Tp + tp)
+B2(TP + i^e-'*3^)
+B3(TP + tp)e-^T"+t^
+£n(rp + *p)e-'(n-1H7>Mp)
(12.65)
436 ADDITIONAL MAINTENANCE POLICES
Since
B1(Tp + tp) = B2(Tp + tp),
= B3(TP + tp),
= Bn(Tp + tp),
+B1(TP + tr)e-alT*+V
+51(Tp + tp)e-,'(n-1)(T')+''))
or
B(TP) = B,(TP + tp)J2 [e-,'(r"+lp)]J . (12.66)
j=o
Since i(Tp + tp) > 0, then e-^T"+^ < 1 and
00
,e-i(TP+tp))j _ I
EXAMPLE 12-5
The net benefit per year obtained from operating a capital equip
ment at age T is given by
6(r) = 32,000e-°O9r,
1. Knowing that
1
tp = 1 month = — year = 0.083 year,
and
i = 10% = 0.10;
-0.09 T „-0.1 T
Br (Tp + tp) = f 32, 000 e dT
Jo
-(15,000- 13,600 e-073 J") e_u''
-0.73 T,A '"
„-0.1 TD
or
Bi(T, + tp) = (168,421)(1 - e-°'19 T*)
-(15,000-13,600 e-073 T") e'01 T".
The total net benefit over a long period of time is then given by
Eq. (12.67), or
_ (168,421)(1 - e-°-19 T") - (15,000 - 13,600 e-°-73Tr)e-ul r?
"V*W ~~ j _ e-0.1(Tp+0.083)
(12.68)
The maximum total net benefit occurs at J^p' = 0. But since
the total net benefit of Eq. (12.68) is a complicated function,
a numerical optimization method should be used. The Golden
Section Method may be used in this case. A computer program
in FORTRAN language has been developed using the Golden
Section Method for this example and is given in Appendix 12 A.
Using this program yields the optimum preventive replacement
age at
T£ = 3.6803 years, or approximately 4 years.
and
*&(v+CR)+ci±a>=0. (»„,
Then,
EXAMPLE 12-6
A system has an exponential times-to- failure pd/with a failure rate
given by
n
where k is the average number of failures per unit time of operation
when n = 1 inspection is made per unit time of operation. The times-
to-restore distribution is exponential with MTTR = - = 24 hr or
0.033 month. The mean time to perform an inspection is j = 8 hr or
0.011 month. The value of the output per month of system operation
is $ 30,000. Costs of repair and inspection per month of system oper
ation are Cr =$250 and Cj = $125, respectively. According to past
experience, the average number of failures per month, when n = 1 in
spection is made, is k = 3. Find the optimal number of inspections per
month, n*, so that the profit per unit time of operation of the system
is maximized.
A(n)=-, (12.73)
n
INSPECTION POLICY FOR MAXIMUM AVAILABILITY 441
n = rib*
It (v+Cr\\
\vtcj). 2
(12.74)
Then,
3(0.033) ^30,000
(30, + 250
n = (12.75)
0.011 V30, 000 + 125
or
n* = 3.006, or n* K 3.
Substituting n* = 3 into Eq. (12.73) yields
3
A(3) = - = 1 fr/month.
/ T f(T) dT
1 - R(Ti)
Thus,
the expected uptime per cycle = Tup,
=[(inspection interval)
X (probability that no failure is detected at inspection)]
-[-[(conditional mean life in one inspection interval)
x (probability that a failure is detected at inspection)],
or
flUTf(T)dT
Tup = Ti R(Ti) + [1 - R(Tj)},
1 - R(Ti)
or
EXAMPLE 12-7
The times to failure of an equipment are normally distributed with
a mean of fi = 5 months and standard deviation a — 1 month. The
time required to effect repair, or replacement, is tR = 0.50 month, and
the time required to effect an inspection is tj = 0.25 month. Find
the optimal inspection interval, Tf , to maximize the long-run average
availability of the equipment.
r/*(Tj) + /2(>T/(T)<ff
A(Ti) = (12.80)
Ti + 0.25 + 0.5(1 - fi(T/)]
Since
f(T) = ! 2 V a ) .
y/2i
then,
r
J—oo
T f(T) dT
-r J — OC oo
T,
T ~r=-
v27T O
1 ( T-u\2
— oo -/27r a
a rj 1 .' r -,a\2 (T — fl
V^Jr
J._
444 ADDITIONAL MAINTENANCE POLICES
+M-$
or
£**»-*■• P^-A--1^.
(12.81)
where
$( ) = cd/ of the standardized normal distribution, 7V(0, 1).
Note that
1 - #(77) = $ (12.82)
and
Substituting Eqs. (12.81), (12.82) and (12.83) into Eq. (12.80) yields
27 [l - * (^)W-* (*£*) ~ ± e
A(Tj) =
Tj + 0.25 + 0.5 $ (Izf:£)
or
A(T/) =
0.25 + 2/ + 0.5 $(Iz^)
(12.84)
Substituting \x — 5 months and a = 1 month into Eq. (12.84) yields
r/-K5-r/)^(^)-^e"(ZV1)
MTl) 0.25 + Tj + 0.5 $ (£^s)
INSPECTION POLICY FOR MAXIMUM AVAILABILITY 445
or
r7 + (5 - r/)«(Tj - 5) - -a-
•/2^
e-5(^-5)
A(Tj) =
0.25 + Ti + 0.5$ (T/ - 5)
(12.85)
Table 12.4 lists the availability values for various values of Tj. A
sample calculation for Tj = 5 months is as follows:
Suppose that a system has failed and the question is whether to replace
it as a whole, or to repair it by replacing only the failed components.
For example, if a car has failed and the cause of the failure is the
transmission, then the owner of the car has to decide whether to buy
a new car or to replace only the transmission. The replacement of a
failed transmission with a new one is considered to be a repair of the
system.
The replacement policies for such systems depend on:
2. the remaining life of the old system if it was repaired at the time
of failure,
/ recommend the type of repair which will give the customer the most
for his money." This is common sense which one would anticipate
ordinary individuals to use.
Ebrahimi [16] formulated the above idea and made it operational.
He proposed two types of policies:
2. A global policy.
Mathematical properties of these policies are developed and some ex
amples show that the two policies can yield different decisions under
the same circumstances.
is greater than a threshold level. The usual models for inspection, peri
odic replacement, block replacement, parallel systems, and cumulative
damage can be transformed into replacement models with threshold
levels. The mean cost-rate of each model is obtained, using well-known
results of reliability theory. The optimum replacement time which min
imizes the cost-rate of an inspection model, is discussed and shown to
exist uniquely.
5. Planning period.
estimated for the wear-life distribution families using the "3a" theo
rem for the normal distribution and the matching percentiles method
for the Weibull distribution. Wear reliability prediction procedures are
developed for different cases using the normal and the Weibull distri
butions. The preventive replacement policy models are developed for
the specified in-service reliability and for the minimum cost. Numer
ical examples are given and discussed. The methodologies presented
can be applied to other failure modes exhibiting cumulative damage
behavior, such as metal fatigue, fatigue crack growth, corrosion, ero
sion, creep, deteriorating material properties in plastics with time, and
so on.
12-1. What is the relationship betwen the Optional and the Age, the
Block and the Pure Corrective Replacement Policies? List all the
advantages and the disadvantages for each policy and compare
them. Give at least two (2) practical examples for which each
of these policies may be applicable. Make justifications wherever
desirable.
12-2. What is the difference between the Optional Replacement Policy
and the Opportunistic Replacement Policy. Give at least two (2)
examples for which each of these two policies may be applicable.
r
PROBLEMS 451
/(T) = m (iT\9 1
_r
e i,
- = fi 7? = 10 ij.
u
(2) Determine the minimum total replacement cost per unit
time of operation in terms of the scale parameter, 77.
(3) Determine the total replacement cost per unit time of oper
ation without any preventive replacements, in terms of the
scale parameter, T).
(4) Discuss the results.
«r' = ^(P'
r?(9!) Vr/y
The Optional Replacement Policy is to be applied to this com
ponent. The time between preventive replacement opportunities,
Tro, is exponentially distributed with an appearing rate of u. The
cost for preventive replacement is Cv = $25. The cost for correc
tive replacement is Cc = $167. Do the following:
452 ADDITIONAL MAINTENANCE POLICIES
- = 0 J) = 10 77.
u
(2) Determine the minimum total replacement cost per unit
time of operation in terms of the scale parameter, n.
(3) Determine the total replacement cost per unit time of oper
ation without any preventive replacements, in terms of the
scale parameter, 7?.
(4) Discuss the results.
12-6. Rework Problem 12-4, but with a constant time between oppor
tunities, Tro.
12-7. Rework Problem 12-5, but with a constant time between oppor
tunities, TT0.
12-8. Assume N components have a Weibull life distribution with pa
rameters of /3 = 2.0, 7 = 0 and 77 = 150 hr. Find the transfer
rate P\t 2 for the Two-stage Replacement Policy with N\ and A'2
components at Stage 1 and Stage 2, respectively, if
(1) transfer is by age,
and
(2) transfer is at random.
12-9. Assume TV components have a Weibull life distribution with pa
rameters of /3 = 1.0, 7 = 0 and 77 = 150 hr. Find the transfer
rate P\t 2 for the Two-stage Replacement Policy with N\ and 7V2
components at Stage 1 and Stage 2, respectively, if
(1) transfer is by age,
and
(2) transfer is at random.
12-10. There are ten (10) interchangeable hne-replaceable units (LRU's)
in a complex electronic system. According to their failure re
placement cost, they can be divided into the following two groups:
PROBLEMS 453
A(n) = -,
n
where k is the average number of failures per unit time of oper
ation when n=l inspection is made per unit time of operation.
The times-to-restore distribution is exponential with MTTR =
- = 20 hr. The mean time to perform an inspection is j = 6
hr. The value of the output per month of system operation is $
25,000. Costs of repair and inspection per month of system op
eration are Cr =$250 and Cj = $125, respectively. According to
past experience, the average number of failures per month, when
n=l inspection is made, is k = 3. Find the optimal number of
inspections per month, n* , so that the profit per unit time of
operation of the system is maximized.
12-15. A system has an exponential times-to-failure pdf with a failure
rate given by
M.)-i.
REFERENCES 455
REFERENCES
1. Woodman, R. C, "Replacement Policies for Component that
Deteriorates," Operational Research Quarterly, Vol. 18, No. 3,
pp. 267-280, 1967.
2. Mou, Zhi-Zhong and Sun, Feng-Bin, "Multimode Replacement
Policies for Large-Scale Industrial Systems," Proceedings of the
International Symposium on Reliability and Maintainability, Tokyo,
Japan, 1990.
3. Bartholomew, D. J., "Two-stage Replacement Strategies," Oper
ational Research Quaterly, Vol. 14, pp. 71-87, 1963.
4. Naik, M. D. and Nair, K. P. K., "Multi-stage Replacement Strate
gies," Operations Research, pp. 279-290, 1965.
5. Vergin, R. C. and Scriabin, M., "Maintenance Scheduling for
Multi-component Equipment," AIIE Transactions, Vol. 9, No.
3, pp. 297-305, 1977.
456 ADDITIONAL MAINTENANCE POLICES
18. Zheng, X. and Fard, N., "A Maintenance Policy for Repairable
Systems Based on Opportunistic Failure-rate Tolerance," IEEE
Transactions on Reliability, Vol. 40, No. 2, pp. 237-244, 1991.
APPENDIX 12A
COMPUTER PROGRAM FOR EXAMPLE 12-5
c
Q ******************************************************
C * COMPUTER PROGRAM FOR OPTIMUM PREVENTIVE REPLACEMENT*
C * AGE DETERMINATION FOR CAPITAL EQUIPMENT BY *
C * MAXIMIZING THE DISCOUNTED BENIFITS USING *
C * THE GOLDEN SECTION (0.618) METHOD *
C ******************************************************
C
EXTERNAL B
WRITE (*,*)» ENTER THE LOWER LIMIT FOR Tp!'
READ(*,*)A1
WRITE (*,*)' ENTER THE UPPER LIMIT FOR Tp! '
READ(*,*)A2
EP=0. 00001
CALL 0PT(A1,A2,EP,XM,FM,B)
WRITE (* , * ) ' ========================================== '
WRITE(*,*)'THE OPTIMUM PREVENTIVE REPLACEMENT AGE IS:'
WRITE(*,*)'Tp*=',XM
WRITEC*, *)'=========================================='
WRITE(*,*)'THE MAXIMUM TOTAL NET BENIFIT IS:'
WRITE(*,*)'B(Tp*)=',-FM
WRITEO, *)'=========================================='
STOP
END
C
C *******************************************************
C * SUBPROGRAMME FOR OPTIMIZATION USING 0.618 METHOD *
C *******************************************************
C
SUBROUTINE 0PT(A1 ,A2,EP,XM,FM,B)
EXTERNAL B
TU=0.618
10 A3=A2-TU*(A2-A1)
F3=B(A3)
20 A4=A1+TU*(A2-A1)
F4=B(A4)
30 IF(ABS(F4-F3)-EP) 1,2,2
1 IF(ABS(A4-A3)-EP) 3,3,4
3 IF(F3-F4) 7,8,8
7 XM=A3
FM=F3
460 ADDITIONAL MAINTENANCE POLICES
RETURN
8 XM=A4
FM=F4
RETURN
4 A1=A3
A2=A4
GOTO 10
2 IFCF4-F3) 5,6,6
5 A1=A3
A3=A4
F3=F4
GOTO 20
6 A2=A4
A4=A3
F4=F3
A3=A2-TU*(A2-A1)
F3=B(A3)
GOTO 30
END
C
C *******************************************************
C * SUBPROGRAM FOR THE OBJECTIVE FUNCTION *
C * —THE TOTAL DISCOUNTED NET BENIFIT *
C *******************************************************
c
FUNCTION B(X)
Bl=168421*(1.0-EXP(-0.19*X))-( 15000- 13600*EXP(-0.73*X))
1 *EXP(-0.1*X)
B=Bl/(1.0-EXP(-0.1*(X+0.083)))
WRITEO,*) 'Tp==' ,X, 'B(Tp)==' ,B
B=-B
RETURN
END
Chapter 13
OVERHAUL POLICIES
13.1 OVERHAUL
Overhaul is a maintenance activity undertaken at scheduled time in
tervals whose primary purpose is to reduce the number of failures and
prevent equipment from reaching the age at which frequent failures
cause substantial loss of performance. In comparison with the preven
tive maintenance policies, given in Chapters 9, 10, 11 and 12 by which
a component, an equipment or a system is restored to "as good as new"
condition after replacement or repair, an overhaul action is considered
as a restorative action which does not return an equipment to "as good
as new" condition but improves its performance. Figure 13.1 illustrates
the effect of overhaul and repair on the equipment's condition. It may
be seen that there is gradual deterioration of the equipment's perfor
mance over time which ends eventually in the replacement of the whole
equipment.
In contrast to preventive maintenance policies where the criteria
are most often to determine the optimum replacement or repair sched
ule so that the total cost per unit of operating time is minimized, an
overhaul strategy is based on making decisions whether, and when, to
overhaul, as well as the level of overhaul to which equipment is to be
repaired. Overhaul versus replacement and repair is a decision most
often based on minimization of future cost, maximization of future
benefits or reducing the failure rate to acceptable levels. The decision
is usually made at regular time intervals so that the loss of production
due to downtime is minimized. For example, overhaul/replacement de
cision is made once a week, once a month or once a year. To optimize
the overhaul/repair/replacement activities over a sequence of regular
intervals, information about overhaul, repair and replacement costs is
required. Since overhaul may consist of different maintenance activi-
461
462 OVERHAUL
I
Overhoul Repair Overhaul Overhaul Replace
Uptime and maintenance time
T = calendar time,
Td = total downtime (overhaul time),
and
3
s
-C
Early failure
U period
_3
ia
—
'3
Start of
wear-out
T= / R(t) dt = / cm dt,
Jo Jo
R(T0) = e-%,
or
500
11(500) = e vaE,
= e -o.i I
or
11(500) = 0.904837.
T = m Q(T0),
= m [1 - R(T0)],
= 5,000 [1 -/?(500)],
= 5,000 [1 -0.904837],
or
T = 475.815 hr.
with the overhaul schedule, T0, so chosen that the occurrence of wear-
out failures is reduced to a minimum. The Weibull pdf is given by
m-lP=*F »■<**. ; (.3.4)
V \ V J
When /? = 1 this pdf reduces to the exponential distribution, and then
only chance failures occur. Assuming that the time T0 is equal to the
time at which chance failures end and wear-out starts; i.e., the time
at which overhaul becomes mandatory, then the reliability function for
this point, where the occurrence of chance failures ends and wear-out
failures start, can be written as
where
(3C = shape parameter for chance failures (/? = 1),
/3u, = shape parameter for wear-out failures (/? > 1),
77c = scale parameter for chance failures (tjc = mc = ■£-),
T)w = scale parameter for wear-out failures,
7C = location parameter for chance failures,
7„, = location parameter for wear-out failures,
Nc
= subpopulation undergoing chance failures,
and
N
—*r = subpopulation undergoing wear-out failures.
NF = N < 1 — I — e * * * + — e ' iv i
OVERHAUL POLICIES 469
EXAMPLE 13-2
A mixed sample of exponential and Weibullian parts is used in
an equipment. Ninety percent of the parts are exponential with the
following parameters:
0C = 1.0; r)c = 5, 000 hr and jc = 0 hr.
Ten percent of the parts are Weibullian with the following parameters:
(3W = 2.5; r)w = 1, 000 hr and 7W = 0 hr.
If the reliability goal is Rgoal = 0.985, determine the following:
1. The interval between overhauls, T0, so that the reliability for a
mission duration of 50 hr satisfies the reliability goal.
2. The mean time between both scheduled and unscheduled over
hauls, MTBO.
3. If 100 such equipment are in operation, determine the number of
those that will fail prior to the overhaul time, T0.
SOLUTIONS TO EXAMPLE 13-2
1. Using Eq. (13.6) and the given parameters, the reliability of
an equipment for a mission duration of 50 hr, given that the
equipment is overhauled every T0 hours, is given by
(T„ +50 \ (Tn+50\''
0.90 e 1 5.°°° I + 0.10 e v »■<>«> '
ftc,w\-l<>i V
0.90 e~(s^o) +0.10 e"vT^oo)
To determine the value of T0 so that
Rc,w(T0, 50) > Rgoal = 0.985, (13.10)
choose different values of T0 and check whether the inequality of
Eq. (13.10) is satisfied. It is determined that for T0 = 700 hr
^^(700,50) = 0.985. Therefore, the interval between overhauls
is 700 hr.
The mean time between overhauls is given by
700
L dT,
where
H = mean wear life,
T = age, or accumulated operating time since new,
and
a = standard deviation of the life times.
Here the case when only one failure mode can occur is considered;
i.e., only chance failures can occur or only wear-out failures can occur
and not both simultaneously.
OVERHAUL POLICIES 471
and so on.
To reduce the number of failures which are caused by wear-out the
overhaul interval, T0 = Tw, or
T0 = n - z a, (13.13)
should be so chosen that the probability of wear-out failures is small.
For single components the replacement or overhaul time must be
kept at (fi — 4 a) or (fi — 5 a) or in between, to prevent wear-out from
appreciably increasing the failure rate. If large numbers of components
are in a system, this replacement or overhaul time must be further
reduced to (/i — 5 a), or even (/i — 6 a), according to the reliability
requirements.
By a proper choice of the replacement, or overhaul, time T0 = (fi —
z a), wear-out failures can be substantially reduced, or even eliminated.
Then, only chance failures would occur and the probability that the
equipment would fail in operation is drastically reduced.
When only chance failures occur between regular overhauls, the
number of parts of the same kind which will have to be replaced because
of failing prior to regular overhaul time T0 amounts, on the average, to
Nf = N Q(T0). (13.14)
When Q(T0) is small and only chance failures occur [2, p. 200],
then
Q(T0) = ^,
m
and Eq. (13.14) yields
Nf = N £ (13.15)
1 m
472 OVERHAUL DEFINED
EXAMPLE 13-3
An equipment has a mean time to failure of ro = 4, 000 hr, a mean
wear-out life of \i — 1,200 hr and a standard deviation of a = 100
hr. To prevent the wear-out failure from occurring prior to scheduled
overhauls the interval between overhauls, T0, should be determined
from T0 = /x — 4 a. Determine the following:
1. The interval between overhauls, T0, and the reliability of an
equipment for a mission duration of T = T0 assuming that only
chance failures occur.
2. The probability of wear-out failures occurring prior to T0.
3. The probability of chance failures occurring prior to T0.
4. Compare the results of Cases 2 and 3.
5. If 100 such equipment are in operation how many will fail due to
chance failures prior to T0?
6. If the interval between overhauls is extended to T0 = 1,000 hr,
what is the probability of wear-out failures occurring prior to T0?
Compare this result to the result of Case 2.
7. Determine the reliability of an equipment for a mission duration
of T = T0 — 1 , 000 hr and compare this result with the result of
Case 1.
8. If 100 such equipment are in operation, how many will fail due
to chance failures prior to T0 = 1,000 hr?
9. What should be the mean wear-out designed-in life if there is
a requirement that not more than 1% fail due to wear-out, as
suming that a good approximation of the standard deviation is
° = "ft-
SOLUTIONS TO EXAMPLE 13-3
Qc(800) = 1 - #(800),
= 1-0.81873,
or
Qc(800) = 0.18127.
or
R{ 1,000) =0.7788.
Comparing the results of Cases 1 and 7 it can be seen that when
the overhaul interval, T0, is extended from 800 hr to 1,000 hr the
reliability of an equipment for a mission duration of T0 = 1,000
hr decreases from 0.81873 to 0.7788.
8. If there are N = 100 units at the start of an overhaul interval
of T0 = 1,000 hr, then the number of those that will fail due to
chance failures is given by
Nf = N Q( 1,000),
- N [1 - R(T0)}
= 100 [1 -0.7788],
= 22.12,
OVERHAUL POLICIES 475
or
N, = 23.
T0 = \x - z O,
p 10
or
(13.16)
*-»(>-a)-
Solving Eq. (13.16) for /x yields
T0
H = 1 - ^-'
10
_1,000
1 2.326
10
'
or
fi = 1,303.1 hr.
which is the ratio of the expected, or average time, Tre, between sched
uled and unscheduled overhauls to the fraction of overhauls caused by
the actual failure of the system. From Eq. (13.17)
j£*M*. (13,8)
mTrc
In case of an exponential system Eq. (13.17) reduces to
So" R(t) dt
mTrt
Q(Tre) '
ff" e-m dt
Q(Tre) '
■mil— e ■• J
m m Q(Tre)
Q(Tre)
or
mT„ = ra
it is known that for a nonmaintained system its mean life is given by
r°°
I m
m = / R(t) dt.
Jo
However, for a maintained system the average constant failure rate is
given by
1 = Q(Tre)
^ava3 — mrr*
mTrc — ~sr '.
r1rt *■■'■•■
I0T"R(t)dt • (13.19)
Equation (13.17) for mrre and (13.19) for Xavg are valid regardless of
the failure distribution of the components. For example, if a component
fails only because of wear-out and is not preventively replaced after Tre
hours of operation, it will fail with a mean time between failures equal
to its mean wear-out life, m.
OVERHAUL POLICIES 477
V 2 n a Jt
T = the age of the component,
and
(n— 1), depending on the decision made at the starting point, n. The
optimum value function, Sn-i(i, j, . . .), is the optimum future cost at
the (n — l)th stage. The optimization criterion is to find the optimum
decision for which the cost of going from the nth to the (n— l)th stage,
Cn, plus the minimum cost of being at the next stage, Sn-\(i,j, . . .),
is minimum.
In general, the recurrence relation is given by
Sn(i, j, ...)= min {Ci + S&.&J, ...)}, (13.22)
where
d = set of decisions possible at each decision point.
The minimum total expected future cost, Sn(i,j,...), is selected as
the minimum among the costs calculated for each possible decision,
d, at the decision point n. Since the future cost, Sn-i(i, j, . ..), is
not known, a general procedure is to start from the stage where it is
known; i.e., 5o(t,j, ...)i and generate the values of the total expected
future cost backwards for n = 1,2,... using the recurrence Eq. (13.22).
Application of this general idea is presented in the remaining sections
of this chapter.
where
P (*>i) = probability that the system will go from state i
to state j in one period of time if decision d is
made,
and
C°(G,G), times the probability that the system is still in good con
dition after one period of time if it was initially good, p°(G,G), plus
the cost of overhaul if the system is initially good and after one period
of time it is in a failed state, C°(G, F), times the probability that the
system is in a failed state after one period of time if it was initially
good, p°(G,F). The next two terms represent the total expected fu
ture cost with (n — 1) remaining intervals if the system is in a good
condition after one period of time, £n_i(G), times the probability that
the system is still in good condition after one period of time if it was
initially good, p°(G, G) plus the total expected future cost with (n— 1)
remaining intervals if the system is in a failed state after one period
of time, Sn-i{F), times the probability that the system is in a failed
state after one period of time if it was initially good, p°(G,F).
The next two lines are for the case if the decision is to replace.
The first two terms represent the cost of replacement if the system is
initially good and after one period of time it is still in good condi
tion, C^iG^G), times the probability that the system is still in good
condition after one period of time if it was initially good, pflp(G!, G),
and the system is replaced, plus the cost of replacement if the system
is initially good and after one period of time it is in a failed state,
CRp(G,F), times the probability that the system is in a failed state
after one period of time if it was initially good, pRp(G,F), and the
system is replaced. The next two terms represent the total expected
future cost with (n — 1) remaining intervals if the system is in a good
condition after one period of time, Sn-i(G), times the probability that
the system is still in good condition after one period of time if it was
initially good, pRp(G,G), and the system is replaced, plus the total ex
pected future cost with (n — 1) remaining intervals if the system is in
a failed state after one period of time, 5'n_i(F), times the probability
that the system is in a failed state after one period of time if it was
initially good, pRp(G,F), and the system is replaced.
The minimum of the two values on the right side of Eq. (13.25)
determines the best decision, and the minimum total expected future
cost if the system is initially in a good state.
n n- 1
Time
The minimum of the two values on the right side of Eq. (13.26) de
termines the best first decision, and the minimum total expected fu
ture cost with n remaining periods of time to operate if the system
is initially in a failed state. In both Eq. (13.25) and (13.26) the op
timum future cost, Sn-i(j), is not known and to find the minimum
total expected future cost, Sn(i), these two equations should be solved
recursively by starting from the stage where the value of Sn-\(j) is
known. Usually So(i), or the optimum future cost with zero remaining
intervals to operate, is known. Now, using Eqs. (13.25) and (13.26)
the minimum total expected future cost with one remaining interval to
operate, S\(i), can be determined. This value is used again to calculate
S2(i) and going backwards, repeating the same procedure, the values
of Sn(i), or the minimum total expected future costs with n remaining
intervals to operate, are determined.
EXAMPLE 13-4 [3, p. 129]
A complex system is put into operation. It is decided that the
system is inspected at regular one-year intervals and, depending on the
condition of the system, replacement, repair or overhaul is undertaken.
The transition probabilities from State i to State j are given in Table
13.1. Costs of the system's operation for a one-year interval, depending
on the system's condition at the start and at the end of an interval,
OVERHAUL POLICIES 483
So{i) = 0.
Then, using Eq. (13.24), the total expected future cost with one re
maining year of operation, n = 1, is given by
N
5i(i) = min ^C(i,j)pd(z,j) (13.27)
d
From Eqs. (13.32) and (13.34) the following conclusions can be drawn:
If the system is in good condition at the decision point, where 2 years
of operation remain, then the minimum total expected future cost is
achieved if the system is overhauled; i.e., 82(G) = $912.50. If the
system is in a failed state at the decision point, with two years of op
eration to go, then the minimum total expected future cost is achieved
if the system is repaired; i.e., 82(F) = $970.00.
Applying the same procedure for three and four remaining years of
operation the minimum total expected future costs, Sz(i) and S^(i),
respectively, can be calculated. Table 13.3 gives the summary of the
calculations for values of n = 1 to 4, the best decisions to be made at
the start of each period and the minimum total expected future costs
for each value of n. It can be seen from Table 13.3 that if there are four
more years for the system to operate, the best decision is to overhaul
if the system is in good condition at the start of a 4-year time and the
minimum achievable total expected future cost is 84(G) = $1,841.60.
If the system is in a failed state at the start of a 4-year time, the best
decision is to repair and the minimum total expected future cost is
S4(F) = $1,900.30. Table 13.3 also gives the best decision at the start
of any of the subsequent intervals with the corresponding minimum
total expected future cost.
EXAMPLE 13-5 [4, pp. 319-320]
A complex system whose age at present is two years is to be in
spected, and either overhauled or replaced at one-year intervals. The
cost of overhaul, C0(i), the annual cost increase over the annual cost
of a new system, CU/(i), and the salvage value of the replaced system
are given in Table 13.4. The acquisition cost of the system is $30,000.
Determine the sequence of decisions for a five-year period in which the
system should be in operation.
SOLUTION TO EXAMPLE 13-5
Let the variable i denote the age of the system and the variable j
denote the order number of a decision point at the start of each year.
Since the initial age of the system is two years, then, i varies from 1 to
7. If the first decision is with the Order Number 1, then the variable
j varies from 1 to 6 when the operation of the system ends. Let c(i)
be the sum of the overhaul cost and the annual cost increase of the
system of age i, or
1 Overhaul
450.00
G
Repair 970.00
F
2 Overhaul
cfor
Example
of
13-4.
aSummary
lTABLE
13.3
culat-ions 912.50
G
Repair 1,435.50
F
3 Overhaul 1,376.90
G
Repair 1,900.30
F
4 Overhaul 1,841.60
G
intervals,
Remaining
at
start
system be
Dto
ecision made
start
at
ii
of
nterval
n of
the
State
iof
nterval future
cost,
Expected
S„(i),S
488 OPTIMAL OVERHAUL/REPAIR/REPLACE POLICY
System age, i
1 2 3 4 5 6 7
c(i) 5 10 9 15 15 19 18
A(i) 10 14 17 20 22 24 30
S(t,6) -20 -16 -13 -10 -8 -6 0
the system of age i, or the net aquisition cost of the system; i.e.,
A{i) = 2D-Ct(i), (13.36)
where the price of the new system is given in thousands of dollars.
Table 13.5 gives the values of c(i), A(i) and 5(i,6), where S(i,6) is the
optimum value function after the fifth year of system operation; i.e., it
is the negative of the salvage value for the system of age i, or
5(i,6) = -C.(i). (13.37)
Using the recurrence relation of Eq. (13.22) and Eqs. (13.35) and
(13.36), the total expected future cost is given by
_,. ,v . f c(t) + S(i + 1,J + 1) 1 overhaul,
(13.38)
where the top line of Eq. (13.38) refers to the overhaul option and the
bottom line corresponds to the replacement option. If the decision is
to overhaul, then the total expected future cost consists of the cost of
OVERHAUL POLICIES 489
the overhaul and the annual cost increase, c(i), plus the optimum value
function after one year of operation, S(i + l,j+l), where the age of the
system now is (i+ 1) and the Order Number of a decision point is (j"+ 1).
If the decision is to replace, then the total expected future cost consists
of the total acquisition cost, A(i), plus the optimum value function
after one year of operation, 5(1, j + 1), where the age of the system
now is i = 1, since the system was replaced and the Order Number
of a decision point is j + 1. As in Example 13-1, to determine the
minimum total expected future cost of the system, initially of age two
years subjected to the given overhaul/replacement policy, Eq. (13.38)
should be solved recursively starting from the decision point where
S(i,j) is known. Since the values of 5(z,6) are known, determine the
values of S(i, 5) as follows: Using Eq. (13.38), the values given in Table
13.5 and assuming that the system's age at the start of the fifth year
is one year, i = 1, yields
c/i ki • fc(l) + 5(2,6) \
5(1,5) = mm (X(1) +5(1,6)/'
r 5-i6i
= min
\ 10 - 20 / '
or
overhaul,
5(1,5) = min {JJ} replace.
Hence, the minimum future cost is
i
490 OPTIMAL OVEPJJAUL/KEPAIR/REPLACE POLICY
and the best decision to be made at the start of the fifth year is to
replace. The next stage the system will be in is (1,6).
If the system's age at the start of the fifth year is three years, i = 3,
then,
| 9-10 1
= min
\ 17-20/ '
or
cio *\ • / -1 1 overhaul,
5(3,5) -min | _3) ^^
and the best decision to be made at the start of the fifth year, if the
system's age is three years, is to replace. The next stage the system
will be in is (1,6).
If the system's age at start of the fifth year is four years, i = 4,
then,
qu « . Jc(4) + 5(5,6) \
5(4,5)= mm jA(4) + 5(M)j,
■ / 15 " 8 \
= mm\20-20/'
or
and the best decision to be made at the start of the fifth year if the
system age is four years, is to replace. The next stage the system will
be in is (1,6).
OVERHAUL POLICIES 491
Now, consider the fourth decision point or the decision point at the
start of the fourth year, (t,4). If the system's age at the start of the
fourth year is one year, i = 1, then,
5(1,4) = min{105:i61},
or
overhaul,
5(1,4) = min (13.43)
{:!} replace.
Since there is a tie, the best decision at the start of the fourth year, if
the system's age is one year, can be either repair or replacement. The
next stage the system will be in is (2, 5) if the decision is to repair or
(1,5) if the decision is to replace.
If the system's age at the start of the fourth year is two years, i = 2,
then,
or
and the best decision to be made at the start of the fourth year, if the
system's age is two years, is to replace. The next stage the system will
be in is (1,5).
Now, consider the third decision point at the start of the third year,
or the Stage (i, 3).
492 OPTIMAL OVERHAUL/REPAIR/REPLACE POLICY
If the system's age at the start of the third year is one year, i = 1,
then,
= mm■{£!}■
•
or
„.. . . J 81 overhaul,
5(1,3)= mm | 9| re
replace.
Hence, the minimum future cost is
5(1,3) =$8,000, (13.45)
and the best decision to be made at the start of the third year, if the
system's age is one year, is to overhaul. The next stage the system will
be in is (2,4).
If the system's age at the start of the third year is two years, i — 2,
then,
e/o « • / c(2) + 5(3,4) \
5(2,3) =mm|A(2) + s(14)j,
f 10 + 61
= mm{l4-l)'
or
overhaul,
5(2,3) =min{^|
replace.
Hence, the minimum future cost is
5(2,3) =$13,000, (13.46)
and the best decision to be made at the start of the third year, if the
system's age is two years, is to replace. The next stage the system will
be in is (1,4).
Consider the second decision point, or the decision point at the
start of the second year. If the system's age at the start of the second
year is one year; i.e., 1 = 1, then,
c/1 _, . f c(l) + 5(2,3) \
5(1'2)=minU(D + 5(1,3))'
f 5+131
= min
I 10+ 8J '
OVERHAUL POLICIES 493
or
overhaul,
5(l,2)=min(J®} replace.
Hence, the minimum future cost is
5(1,2) =$18,000, (13.47)
and either an overhaul or a replace decision can be made with the same
minimum total expected future cost.
If the system's age at the start of the second year is three years;
i.e., i = 3, then,
5(3,2)
= min{c(3)
minU(3) ++ 5(4'3U
S(l,3)/'
f 9+191
= nun
1 17+ 8J'
or
overhaul,
5(3,2) =minj ^} replace.
Hence, the minimum future cost is
5(3,2) =$25,000, (13.48)
and the best decision to be made at the start of the second year, if the
system's age is three years, is to replace.
The next stage the system will be in is (1,3). What is left is to
consider the initial stage, or the first decision point where the system's
age at the start of the first year is two years; i.e., i = 2. Then,
f 10 + 25 1
= min
\ 14 + 18 J'
or
overhaul,
5(2,l) = min{3^} replace.
Hence, the total minimum expected future cost for a system age of two
years, and five years of operation to go, is
5(2,1) =$32,000, (13.49)
494 OPTIMAL OVERHAUL/REPAIR/REPLACE POLICY
Minimum future
Stage, cost, in $1,000, Next stage,
(m) S(i,j) Decision (hj)
(1,5) -11 0 (2,6)
(2,5) -6 Rp (1,6)
(3,5) -3 Rv (1,6)
(4,5) 0 Rp (1,6)
(6,5) 4 Rp (1,6)
(1,4) - 1 Rp or 0 (1,5) or (2,5)
(2,4) 3 Rp (1,5)
(3,4) 6 Rp (1,5)
(5,4) 11 Rp (1,5)
(1,3) 8 0 (2,4)
(2,3) 13 Rp (1,4)
(4,3) 19 Rp (1,4)
(1,2) 18 RporO (1,3) or (2,3)
(3,2) 25 Rp (1,3)
(2,1) 32 Rp (1,2)
and the best decision to be made at the start of the first year of op
eration is to replace. Since the annual cost increase, which is the
difference between the cost in a particular year and the first year of
operation, is used throughout these calculations instead of the actual
annual cost, then the total future cost of $32,000 is the additional cost
for a five-year operation and failure costs determined on the cost of
the first year. Table 13.6 gives all stages necessary to determine the
minimum total expected future cost and the sequence of decisions for
the next five years of operation. To determine the sequence of best
decisions, which results in the minimum total expected future cost,
consider Table 13.6. Start from the initial stage where the system's
age is two years, or (2, 1). The best decision that minimizes the total
future cost, which is that of all future stages, is to replace and the next
stage is (1,2). At stage (1,2) either overhaul or replacement can be
chosen with the same future cost and the next stage is either (1,3) if
replacement, or (2, 3) if overhaul is chosen. If overhaul is chosen the
OVERHAUL POLICIES 495
next stage is (1, 3). The minimum future cost at stage (1,3) is obtained
if the overhaul decision is made and the next stage is (2,4). At stage
(2, 4) the replacement decision minimizes the future cost and leads to
the stage (1,5). The final decision at the start of the fifth year is over
haul and after the fifth year of operation the system's function ends.
Since at stages (1,2) and (1,4) both replacement and overhaul deci
sions are possible, then there are three possible sequences of decisions
which have the same minimum total expected future cost; i.e.,
1. replace, replace, overhaul, replace, overhaul, dispose,
or
2. replace, overhaul, replace, replace, overhaul, dispose,
or
3. replace, overhaul, replace, overhaul, replace, dispose.
Each of these decisions is made at the beginning of a decision interval
and after the fifth year of operation the system is disposed.
or
AT W
n g + u(i) = mjn \ £ Cd(i, j) pd(i, ;) + X>d(*> J') (n " 1) 9
j=l 3=1
N |
y£JPd{hJ) v(j) \
+
N .
Since £ Tr (*»i) = 1> then,
(13.54)
EXAMPLE 13-6
The system of Example 13-4 is to be maintained over "a long"
period of time. The transition probabilities are given in Table 13.1 and
the operation costs in Table 13.2. Determine the best overhaul/repair/
replace decisions or the optimum maintenance policy and the steady
state average cost per unit time of operation.
SOLUTION TO EXAMPLE 13-6
To determine the optimum decisions depending on the state of the
system at start of one-year interval use the Howard's algorithm as
follows:
Step 1
Assume the following decisions at the start of an interval:
Step 2
Since the variable i can be only in N = 2 states; i.e., i = G or F,
then using Step 2 of the Howard's algorithm yields
v(F) = 0. (13.55)
Step 3
;
Using the data given in Tables 13.1 and 13.2 and the decisions
defined in Step 1, solve Eq. (13.53) for the steady state average cost
per interval, g, and the transient cost, v(i), or for i = G and for j = G
and F,
g + v(G) = CR'(G,G)pR''(G,G) + CR*(G,F) P^(G,F)
+pR*{G,G) v(G) +pR*(G,F) v(F), (13.56)
498 INFINITE TIME HORIZON
Using the data from Tables 13.1 and 13.2, and Eqs. (13.55) and (13.56)
yields
g + v(G) = (500)(0.95) + (1,500)(0.05)
+(0.95) v(G) + (0.05) (0),
and Eq. (13.57) yields
g + 0 = (500)(0.95) + (1,500) (0.05)
+(0.95) v(G) + (0.05) (0),
or
g + V(G) = 475 + 75 + (0.95) v(G),
and
g = 475 + 75 + (0.95) v(G),
or
g + v(G) = 550+ (0.95) v(G), (13.58)
and
g = 550 + (0.95) v(G). (13.59)
The solution to the system of Eqs. (13.58) and (13.59) is
g = 550 (13.60)
and
v{G)=0. (13.61)
Step 4
Using Eqs. (13.55) and (13.61) determine the best decisions for
each state of the variable i which minimizes Eq. (13.54).
If the system is initially in good condition, or i = G, using Tables
13.1 and 13.2, j = G and F, and Eq.(13.54), yields
or
450 overhaul,
min
550 replace.
Hence, the minimum of Eq. (13.54) is obtained if the decision is to
overhaul.
If the system is initially in a failed state, or i = F, using Tables
13.1 and 13.2, Eq. (13.54) and j = G and F, yields
ZCr(F,j)pr(FJ)+Y:pr(F,j)v(j) repair,
j=\ j=l
nun
£ C"*(FJ) pR'(FJ) + E PR'(F,j) v(j) replace,
.7=1 j=l
(100)(0.60) + (1, 100)(0.40) + (0.60)(0) + (0.40)(0)
min
(500)(0.95) + (1,500)(0.05) + (0.95)(0) + (0.05)(0) J '
or
500 repair,
min
550 replace.
Hence, the minimum of Eq. (13.54) is obtained if the decision is to
repair.
The first iteration results in the new set of decisions; i.e.,
Step 5 -» Go to Step 3
Using data from Tables 13.1 and 13.2, and Eqs. (13.55) and (13.62)
yields
g + v(G) = (200)(0.75) + (1,200)(0.25)
+(0.75) v(G) + (0.25)(0),
and Eqs. (13.55) and (13.63) yield
Using Eqs. (13.55), (13.66) and (13.67), or the values of v(F), v{G)
and g obtained in Step 5, determine the best decisions for each state
of the variable i which minimizes Eq. (13.54).
If the system is initially in good condition, or i = G, and j = G
and F, then
Zi = K*fiix)dX- i»:iT(»)
Jf* fi(x) dx
Using the recurrence relation of Eq. (13.24), the total minimum ex
pected future cost for an equipment of a given age, z, subjected to an
overhaul/replace maintenance policy, with n remaining time intervals
of operation, is given by
Sn(i) = mm{Cn(i,j) + 5„_i(j)}, (13.71)
Li
where
Cn('»i) = expected cost of the first decision for an equipment of
age i and with n remaining time intervals of operation,
and
Sn-i(j) = minimum expected future cost for (n — 1) remaining
time intervals of operation.
The expected cost of the first decision, Cn(i, j), is the sum of the
expected cost of overhaul times the probability that the overhaul cost
is less than the overhaul cost limit, Lj, and the cost of a replacement,
A*, times the probability that the overhaul cost exceeds the overhaul
cost limit, Lj. Using Eqs. (13.69) and (13.70) the expected cost of the
first decision is given by
Cn(i,j) =U Pi(Li) +A*[1- Pi(Li)}. (13.72)
OVERHAUL POLICIES 503
Then,
Cn(iJ) = £glM± fU fi{x) d* + A* [l - [Li /<(*) dx
/o ' Mx) dx Jo [ Jo
or
or
Sn(i) = min fZi Pi{Li) + A* [1 - ^(£,)]
+Sn.l(i + 1) Pi(Li) + 5n_!(l) [1 - Pi(Li)]} . (13.75)
Apparently, the starting condition, or the minimum expected future
cost for n = 0 remaining time intervals of operation, is
S0(i) = 0 for all i. (13.76)
The optimum overhaul cost limits are so determined that the total ex
pected future cost given by Eq. (13.75) is minimized. The iteration
process starts from the terminating stage; i.e., n = 0, where the total
expected future cost is known, and going backwards the minimum ex
pected future costs, Si(i), Si(i), • • -, are determined for possible equip
ment ages, t.
Alternatively, the optimum overhaul cost limits can be determined
as follows: Assume that the equipment is in state n, with the optimum
value function Sn(i), and the overhaul cost is $i. If the equipment is
overhauled, the total cost is given by
x + Sn-i(i + 1), (13.77)
504 OPTIMAL OVERHAUL COST LIMITS
or
or
S^t) = Li Pi(Li) + A*[l - Pi(Li)].
For the three-year-old unit and one remaining year of unit operation,
replacement is compulsory, or
5i(3) = $150. (13.81)
For the two-year-old unit and one remaining year of unit operation,
since the overhaul cost is uniformly distributed over the range (0; $150)
and the cost of a new unit is A* = $150, overhaul should be always
done. Consequently, P2(-^2) = 1, because overhaul is a certainty. The
overhaul cost pdf at the end of the second year of a unit's operation is
Mx) = 150
> 0u _
< x < 150. (13.82)
or
-i x Ln
150
dx,
'
l50|0
or
P2(L2)
v ' =—
150 . (13.87)
_ Jo x 150 °x
Lo =
2L2'
or
u —T (13.88)
Using Eqs. (13.75) and (13.89), the total expected future cost for the
unit of age two years, and two remaining years of operation, is given
OVERHAUL POLICIES 507
by
= mm
a + 150
l2 300
+150^ + 50
or
52(2)=tn{i+200(1-i)+L2}- (13.90)
To find the optimum overhaul cost limit, L2, which minimizes the total
expected future cost given by Eq. (13.90), find the first derivative of
52(2) with respect to L2, equate it to zero and solve for L2. Then,
dS2{2) _ 212 200
dL2 300 150
or
d52(2) ^ U_ _50_
dL2 150 150'
Now,
L2 50
= 0,
150 ~ 150
and
L2 = $50. (13.91)
Substituting Eq. (13.91) into Eq. (13.90) the minimum total expected
future cost is
502
52(2) = |_ + 200(l-^)+50,
2,500 100
300 +200l50+50'
or
52(2) = $191.67. (13.92)
_.!._
508 OPTIMAL OVERHAUL COST LIMITS
Hence, if the estimated overhaul cost is less than $50, then the decision
is to overhaul, otherwise replace the unit. The minimum total expected
future cost for the unit of two years of age, and for two remaining years
of operation, is $191.67.
For a one-year-old unit and two remaining years of operation, the
probability that the overhaul cost is less than the overhaul cost limit,
is given by
= r(H+150(1-^)
+5,(2)^ + 5,(1) (l- j±)}. (13.«)
Substituting the values of Si (2) and Si(l) from Eqs. (13.83) and
(13.85), respectively, into Eq. (13.95) yields
-2
OVERHAUL POLICIES 509
or
I, = 100-,
4
or
Lx = $125. (13.97)
Since the optimal cost limit, L\ = $125, is greater than the overhaul
cost whose range is (0, $100) for a unit age of one year, the decision
should always be to overhaul. Then, the minimum total expected
future cost for the unit with an age of one year and two remaining
years of operation is given by
52(1) = Z1+51(2),
Li
-L x Jx{x) dz + 5i(2),
-L V2
x
100
100
dx + 75,
'
+75,
2(100)
= 50 + 75,
or
52(1) = $125. (13.98)
Hence, the optimum total expected future cost for the unit of an age
of one year and for two remaining years of operation is 52(1) = $125,
and the decision should be to overhaul regardless of the overhaul cost
estimate.
Stage 3, n = 3
Since the unit is initially new at this stage, when three years of
unit operation remain, the system can not be three years of age.
510 OPTIMAL OVERHAUL COST LIMITS
Therefore, the minimum total expected future cost, S3(3), should not
be considered.
Consider a unit of two years of age, or i = 2, and three remaining
years of unit operation, or n = 3. Using Eqs. (13.75), (13.86), (13.87),
(13.88) and (13.98) the total expected future cost is given by
S3(2) = min {Z2 P2{L2) + A* [1 - P2(L2))
+S2(3) P2(L2) + S2(l) [1 - P2(L2)}} ,
. (L2 L2 ( L2\\
— mm { — —— + 150 1 >
L2 I 2 150 V 150/ J
or
«"S+»(,-ra)}'
or
53(2) = $256.25. (13.102)
Consider a unit of one year of age, i = 1 , and three remaining years
of operation, or n = 3. Using Eqs. (13.75), (13.93), (13.94), (13.92)
and (13.98) yields
53(1) = min{l, PiiL,) + A* [1 - P^)}
+S2(2)Pl(L1) + 52(1)[1-P1(Z1)]},
mir
Li
+m.67iL + 125(,-^L)},
OVERHAUL POLICIES 511
or
_ ... = ^^r-
S3(l) 83.332 + nrwm /
275 [1 83.33
(l-^)+0-M67)(83.33),
200
or
53(1) = $240.28. (13.106)
The minimum total expected future cost for the unit of one year of age
and three remaining years of operation is 53(1) = $240.28.
Stage 4, n = 4
Since the unit is new at the start of the five-year period of operation,
at Stage 4 where four years of operation remain, the unit's age can be
only one year, then only one state should be considered; i.e., n = 4
and i — 1.
Using Eq. (13.80) the optimum overhaul cost limit is then given
by
Lx = A' + S3{1) - S3(2). (13.107)
Substituting Eqs. (13.106) and (13.102) into Eq. (13.107) yields
Lx = 150 + 240.28 - 256.25,
or
Lx = $134.03. (13.108)
512 OPTIMAL OVERHAUL COST LIMITS
Since the overhaul limit is greater than the maximum overhaul cost
for the unit of age of one year, then the unit should be overhauled
regardless of the cost estimate. From Eq. (13.75) the minimum total
expected future cost for the unit with an age of one year, and four
years of operation remaining, is given by
54(1) =Xi+53(2),
= 50 + 256.25,
or
Rp
S2(3) -S,(l) » S0(2)
O
S4(D ►S3(2)
O O
S2(D ^S,(2) -S0(3)
System age, i
1 2 3 4 5
Overhaul cost range,
(1,2) (2,4) (3,6) (4,8) (5, 10)
Salvage value,
S(i) 3 2 0 0 0
514 OPTIMAL OVERHAUL COST LIMITS
Determine the optimum overhaul cost limits so that the total expected
future cost is minimized.
SOLUTION TO EXAMPLE 13-8
The system is to be in operation for four more years, therefore
the variable n, number of remaining years of operation, has the values
n = 0, 1,- • • ,4. Since the overhaul/replace policy is to be applied at
a time when the system's age is one year, the system's age, i, has
the values t = 1,2,***,5. To determine the optimum overhaul cost
limits for each interval under consideration, Eq. (13.75) should be
used starting from the terminating stage, or n = 0, when the system's
operation ends. Therefore, the values of the minimum total expected
future cost, So(i), for all possible system ages should be determined
first. Because with So(i) we are reaching the end of operation the
only cost left is the salvage value which is obtained from Table 13.8.
Consequently,
So(l) = -3, 50(2) = -2,
and
50(3) = S0(4) = 50(5) = 0.
Since the overhaul cost is uniformly distributed in each age interval,
the probability that the estimated cost will be less than the overhaul
cost limit, Lj, is given by
{0 for Li < a,
^ for a < Li </?, (13.110)
1 for Li > /?,
and the expected cost of overhaul and replacement is given by
x
^ fa' ih **
U =
tirfc**'
Li
X2
a
ET'
2x
a
L\-c?
2 (U -a)'
or
Li + a
Li = -^—. (13.111)
OVERHAUL POLICIES 515
Then,
0 for Li < a,
Li = } ^2 for a<Li<0, (13.112)
2±£ for Li > /?.
Stage 1, n = 1, i - 1,2,3,4
Using Eq. (13.80) the optimum overhaul cost limit for a system
age varying from t = 1 to 4 years yields
Li = A*- S(i) + Sn_!(l) - Sn-i(i + 1),
L, = A* - S(l) + So(l) - So(2),
= 10-3-3 + 2,
or
Li =6. (13.113)
L2 = ^-5(2)+5o(l)-50(3),
= 10-2 + 3-0,
or
L2 = 5. (13.114)
L3 = ^-5(3) + 5o(l)-50(4),
= 10-0-3-0,
or
18 = 7, (13.115)
and
L4 = A* - 5(4) + 50(1) - 50(5),
= 10-0-3-0,
or
L4 = 7. (13.116)
If i = 1, then the overhaul cost is uniformly distributed in the range
(a, (3) = (1,2). Since the optimum overhaul limit is L\ = 6, then the
estimated overhaul cost is always less than L\ and an overhaul decision
51 6 OPTIMAL OVERHA UL COST LIMITS
1+2
2 '
or
Ii ■ 1.5. (13.117)
Similarly, for i = 2, (a, (3) = (2,4), and since Li — 5 then Pi{L<}) = 1.
Then,
-=, 2+4
or
Z2 = 3. (13.118)
For i = 3, (a,/?) = (3,6), and since L3 = 7 then P3(L3) = 1. Then,
-=-3+6
u -—.
or
L3 = 4.5. (13.119)
For i = 4, (a,/?) = (4,8), and since L4 = 7, then, from Eq. (13.111),
3
~ 4'
or
2i
" 2'
or
I4 = 5.5. (13.121)
OVERHAUL POLICIES 517
Using Eq. (13.80) the optimum overhaul cost limit for a system
age varying from i = 1 to 3 years yields
Li = A* - s(i) + S„_,(l) - 5„_i(i + 1),
L, =A*-s(l) + Si(l)-Si(2)1
= 10-3-0.5-3,
or
L, = 3.5. (13.126)
L2 = ^-*(2) + 5i(l)-5,(3),
= 10 - 2 - 0.5 - 4.5,
or
L2 = 3, (13.127)
and
L3 = A' - s(Z) + Si(l) - Si(A),
= 10 - 0 - 0.5 - 5.875,
or
L3 = 3.625. (13.128)
OVERHAUL POLICIES 519
1+2
2 '
or
Lx = $1.5. (13.129)
For t = 2, (a,/?) = (2,4) and L2 = 3, then
Li — a
P2(L2) -
0-a'
3-2
4-2'
or
P2(Z2) = 0.5, (13.130)
and
-p 3+2
or
L2 = 2.5. (13.131)
For i = 3, (a,/?) = (3,6) and L3 = 3.625, then
L3 - a
Ws) = /?-a'
3.625 - 3
6-3 '
or
P3(L3) = 0.21, (13.132)
and
T L3+a
3.625 + 3
or
L3 = 3.313. (13.133)
520 OPTIMAL OVERHAUL COST LIMITS
Using Eq. (13.80) the optimum overhaul cost limit for the system
of ages i = 1 and 2 years, yields
Li = A* - S(i) + S„_,(l) - 5„_!(i + 1),
L1 =A'-S(l)+Si(l)-Si(2),
= 10-3 + 4.5-7.25,
OVERHAUL POLICIES 521
or
Ii = 4.25, (13.137)
and
L2 = A*-S(2) + S2(1)-S2(3),
= 10 - 2 + 4.5 - 9.434,
or
L2 = 3.066. (13.138)
If i = 1, (a,/?) = (1,2) and Lx = 4.25, then,
Pi(Li) = l, (13.139)
and
T a+P
Ll== 2 '
1+2
2 '
or
L, = 1.5. (13.140)
If i = 2, (a,/?) = (2,4) and L2 = 3.066, then,
L2 — a
P2(L2) =
/3-q'
3.066 - 2
4-2 '
522 OPTIMAL OVERHAUL COST LIMITS
or
Using Eq. (13.80) the optimum overhaul cost limit for the system
of age i = 1 year, yields
L, =i4*-5(l) + S8(l)-Si(2)1
= 10-3 + 8.725- 12.216,
OVERHAUL POLICIES 523
or
Li = 3.509. (13.145)
Since the overhaul cost limit is greater than the maximum overhaul cost
for a system of one year of age, then the overhaul should be undertaken
regardless of the overhaul cost estimate. Then, P\{L{) = 1 and
a+p
u= 2
14-2
or
Li = 1.5, (13.146)
and
S4(l) = LX F\(L{) + [A* - .(1)][1 - Pi(Li)]
+ S3(2)P1(L1) + S3(l)[l-Pi(L,)],
= (1.5)(l) + (10-3)(0)
+ (12.216)(1) + (8.725)(0),
or
54(1) = 13.716. (13.147)
Hence the total expected future cost for the system of an age of one
year and four remaining years of operation is $13,716, which may be
achieved using the optimal overhaul cost limits given in Table 13.12.
The word "overhaul" means that the overhaul decision should be made
regardless of the overhaul cost estimate since it is always less than the
overhaul cost limit. Figure 13.5 gives the optimal paths for this system
which is initially one year old, and is to be maintained over the four-
year period with the optimal overhaul cost limits given in Table 13.12.
524 OPTIMAL OVERHAUL COST LIMITS
o o o o
S4(l) —S3(2) <3064'S2(3) <3.625 ' S.W <7 'So(5)
S,(l) S0(l)
S0(2)
S0(3)
PROBLEMS
13-1. 1,000 units with the mean time to failure of m = 1,000 hr, are
to be overhauled at regularly scheduled time intervals, T0, equal
to 1,000 hours. Determine the following:
U-
Xswo= 10 fir/10 hr
A.SWQ= 50fr/106hr
^SWE= 100 fr/106 cycles
Fig. 13.6 - Standby system with one active and one standby
unit for Problem 13-4.
528 OVERHAUL POLICIES
Jl,E=100fr/10 hr
X2E=120fr/106hr
A*,= 0
XsE= 10fr/106hr
U-
"•swo — 0
"•SWQ — 0
kSWE = 50fr/10 cycles
Fig. 13.7 - Standby system with one active and one standby
unit for Problem 13-5.
REFERENCES
SPARES PROVISIONING
where
Nf{Tz — T\ ) = estimate of the expected number of failures
for a life period (T2 - T\), per part,
A(T) = failure rate function for a life period [T2 — T\)
for that part,
and
(22 — T\) = part life period for which the spare parts pro
visioning is to be determined.
Equation (14.1) applies to any type of reliability bathtub curve (RBTC),
with or without preventive and/or corrective maintenance, provided
the RBTC is representative of the situation for which the number of
spare parts is determined, the appropriate A(T) is used, and the total
number of identical parts in use remains essentially constant. Then,
the total number of expected spare parts is given by
NFT(T2 -Tl) = NTx NF(T2 - r,), (14.2)
533
J__
534 SPARES PROVISIONING
where
Nrrpi-Ti) = total number of expected spare parts in life
period (T2 - Tx),
and
Nt - total number of identical parts in use during life period
(Ta-Tj).
If N7 • varies with T, then
where
-2 t" \
= 1.
Then,
x -x
+ -e * e -x
x + — x
x + ■■■+—re -X
x + ■ ■ ■ = 1.
12 n!
SPARES PROVISIONING 535
Each term represents a probability, and the sum of all these prob
abilities is equal to 1. Hence, each term is a term of a pdf, and in this
case of the Poisson pdf.
The interpretation of this distribution is as follows: If x is taken to
be the expected, or average, number of occurrences of an event, then
e~x = probability that that event will not occur if x remains
constant,
xe~x = probability that that event will occur exactly once,
x2 x = probability that that event will occur exactly twice,
—e
and so on.
In reliability, the event of concern is failure, and the average number
of failures in time t is given by x — At, when A is constant. Conse
quently, e~x = e~Xt, which is R(t) for a single system having a constant
failure rate, A, gives the probability that no failure will occur in time
t. xe~x = Aie~Al is the probability of exactly one failure occurring
in time t, [i2/2!]e-1 = [(At)2/2!]e_A< is the probability of exactly two
failures occurring in time t, and so on.
Therefore, the probability of exactly k failures occurring in t is
given by
/(2) = e-^2
2! '
1(2) = e- o.ois (0-015)2
2 '
„ , . ,0.000225
/(2) = (0.9851) ,
or
/(2)= 0.00011,
a much lower probability than for one failure.
SPARES PROVISIONING 537
4. The probability that two or fewer failures occur during this mis
sion is given by
EXAMPLE 14-2
In a system there exists a very critical unit which requires spares
to attain a specified unit reliability of 99%, for a period of 250 hr. The
unit has an MTBF of 1,250 hr and exhibits a constant failure rate
characteristic.
How many spares would be required if the unit is easily accessible
and can be replaced almost immediately, by successfully plugging in an
identical spare when the functioning unit fails, to increase its reliability
from 81.87% to 99%?
SOLUTION TO EXAMPLE 14-2
The solution may be found by using the Poisson distribution and
answering the question, "How many failures, equal to the number of
spares, can be tolerated to attain the reliability of 99%?" or the ques
tion, "How many standby (spare) units are required to attain the
reliability of 99%?" Therefore,
m -£<-**#.i=o
Xt (Xt)2 + ••• + (At)*
F(k) = e -Xt 1 + T7 +
1! 2! fc!
and
Ai=(M7b)(,»=(l^)(250) = 0-2-
Consequently,
This equation should be solved for the nearest integer, k, which satisfies
the equality; then the required number of spares is found to be
k-2.
With two spare units, the actual reliability is
Rs = F(2) = 0.99885,
whereas with one spare it would be
Rs = F(l) = 0.98248.
Consequently, k = 2 is the right answer.
EXAMPLE 14-3
A battery has an expected failure rate of 0.01 fr/hr and is used 24
hr per day.
1. How many spares will be required for a three-calendar-month
period (assume 30 days per month) for a 95% probability (ad
equacy, assurance, or confidence) that there will be a sufficient
number of spares?
2. What would the battery reliability be for a 24-hr period?
3. If a battery adequacy of 95% is required for a 24-hr period, how
many spares would be required for a three-calendar-month pe
riod, assuming the replacement of the failed batteries is immedi
ate?
4. Compare and discuss the results obtained in Cases 1 and 3.
t = 2,160 hr,
SPARES PROVISIONING 539
and
A = 0.01 fr/hr.
Then,
TTsp = (0.01X2, 160),
or
NSp = 21.6 failures.
For a 95% assurance of having sufficient spares
'h >'■ i
or
R(t) = 0.7866.
1 + ^p+%£ +
1! 2!
For Ns = 1,
0.95 < e-024(l + 0.24),
0.95 < 0.9754;
therefore, one spare will be required for a 24-hr operating period.
Extending this to a three-calendar-month period, or to 90 days,
yields
Ns = (1)(90) = 90 spares.
540 SPARES PROVISIONING
EXAMPLE 14-4
There are N = 100 identical units operating in the field, or in a
production facility. Each unit has a mean time to failure, MTTF , of
500 hr. For a period of 6 months, during which each unit operates
cumulatively t0 = 600 hr, determine the following:
1. The average number of spare units required.
2. The number of spare units required at the 80%, 90%, 95% and
99% confidence levels.
SOLUTIONS TO EXAMPLE 14-4
Actual
Confidence Number confidence
level, of level,
% spares %
80 129 80.8
90 134 90.5
95 138 95.2
99 146 99.1
For example, for CL — 80% the closest value to it, in the C(Ns)
column in Table 14.1, is 0.80817221 in decimals. Going to the
corresponding value in the first column yields Ns = 129, or nine
(9) more than the average. Going to the value of C(Ns) closest
to CL = 90% yields Ns = 134, or 14 more than the average.
For CL = 95% Table 14.1 yields Ns = 138, or 18 more than the
average, and for CL = 99% it yields Ns = 146, or 26 more than
the average. The number of spare units required at the desired
confidence levels are summarized in Table 14.2. Reference [3]
may also be used to find Ns-
or
The number of units that will fail, Nj?, and need to be spared would
be given by
NF = NTQ{t) = NT[1-R(t)]. (14.7)
The conditional reliability function, R(T,t), may be used to find
the number of units which should start the new mission, starting at
the age T, Ns(T), to end up with the desired number at the end of the
new mission of t duration, Ns(T + t), and the number that should start
at age zero, Ns(0), to end up with desired number at the end of the
new mission, Ns{T + t), and vice versa. By definition the conditional
reliability is given by
Also, by definition,
NS(T + 1)
R{T + t) =
Ns(0)
Therefore, the number of units that should start at age zero, Ns(0), is
given by
Ns(T + t)
Ns(0) = (14.11)
R(T + 1) '
EXAMPLE 14-5
The times-to-failure distribution of identical units is represented
well by the normal distribution with mean T = 43,679 hr and a stan
dard deviation of aj — 562 hr. Do the following:
1. Find the number of units which should start the new mission
of 300-hr duration, each unit having already accumulated 42,850
hr, to end up with 100 such units at the end of the mission.
2. Find the number of units which should start at age zero to end
up with 100 such units at the end of the new mission.
where
or
FromEq. (14.11)
s{) R(T + t)
0.8267'
or
7Vs(0) = 121 units.
Consequently, not 100 but 121 units need to be provided, or
so to speak 21 additional spares should have been on hand to
accommodate this situation.
hours of operation is R(T\), and from Eq. (14.7) the number that will
fail by T\ hours of operation is
NF-r(Tx) = No Q(Tx) = JV0 [1 - R(Ti)]. (14.13)
These are replaced by fresh ones and they operate T hours thereafter.
The number of these replaced units that will fail after T additional
hours of operation, using Eq. (14.13), would be
NF.R(T) = JV>_„(r,) Q(T) = N0 [1 - JJ(I\)][1 - R(T)}. (14.14)
The number of those that do not fail by T\ is
Ns(T1) = NQR(Tl). . (14.15)
The number of these units that will fail while operating T additional
hours, using Eq. (14.15), would be
Nf-nbHTuT) = NsiTx) [1 - R(TUT)),
or
g(7j + T)
NF-nk(TuT) = NoR(Ti) l (14.16)
R(TX)
Consequently, the total number of such units that will fail by (T\ + T)
hours of operation, under the condition that those that fail by T\ hours
are replaced, is given by the sum of Eqs. (14.13), (14.14) and (14.16),
or
NFMTi + T) = No [l-UW)]
+N0[l-R(T1)][l-R(T)\
R{TX + T)
+N0 R(T,) 1- (14.17)
m)
Simplification of Eq. (14.17) yields
NfMTi + T) = No [2 - R(T,) - R(T)
+fl(T1)fl(T)-JR(T1+T)]. (14.18)
Eqeuation (14.18) gives the average number of spares that should be
provided for A^o such units with the replacement policy considered here.
Of those that fail by T\ hours of operation and are replaced, the
number that will survive after additional T hours of operation is given
by
NS-r(T) = Nf-r(Ti) R(T) = N0 [1- R{t\)] R(T). (14.19)
548 SPARES PROVISIONING
or
Ns-nr{Tx + T) = N0 R(T, + T). (14.20)
Out of the iVo that started out, the number of those that survive, after
replacement of those that failed by T\, is given by the sum of Eqs.
(14.19) and (14.20), or
Ns-rM +T) = N0 {[1 - RiTi)} R(T) + fi(T, + T)}. (14.21)
EXAMPLE 14-6
Identical types of bearings have a Weibull times-to-failure distribu
tion with the following parameters:
0 = 2, r\ - 2, 000 hr, and 7 = 0 hr.
Do the following:
1. If 100 of such bearings are operating in identical equipment at
the same application and operation stress level, how many will
survive if each operates T\ — 1, 300 hr at 675 rpm, and how many
will fail?
2. If the failed bearings are replaced by new ones and all of them
operate for T — 700 hr thereafter, how many of the fresh ones
will fail?
3. Of those that survived T\ hours, how many will fail during the
additional T hours of operation?
4. What is the total number of bearings that will fail by (7\ + T)
hours of operation, given that 100 start at age zero and that
those that fail by T\ hours are replaced by fresh ones?
5. What is the total number of bearings that survive after (7\ + T)
hours when we follow the policy of replacing the failed ones by
T\ hours?
SPARES PROVISIONING 549
, 1 ,300 \2
i2(l,300hr) = e~{™°°> ,
or
#(1,300 hr) = 0.655406.
The number of bearings that will survive is
Ns(Tr) = JVs(l,300 hr) = NoR^) = 100 x 0.655406,
= 65.5406,
or
-/V5(l,300hr)2 65.
The number of such bearings that will fail is
Nf-r^) = N0 Q(7\) = N0 [1- «Ct|)J,
= 100 x (1 -0.655406),
= 34.4594,
or
iVF_fl(l,300hr)^35.
2. From Eq. (14.14)
nf.r{t) = No [i - «(r,)][i - j?(r)],
where
/ 700 \2
i2(T) = e (w*> ,
or
72(700 hr) = 0.884706.
Then,
NF.R(T - 700 hr) = 100(1 - 0.655406)(1 - 0.884706),
= 100 (0.344594)(0.115294)= 3.97,
or
NF-.R(700 hr) S 4.
Therefore, four out of the fresh bearings will fail while operating
700 hr after replacement.
550 SPARES PROVISIONING
Then
or
or
or
or
A^5-fl(2,000hr) = 67.
r
Jo
R{t) dr
552 SPARES PROVISIONING
gives the mean time between both preventive replacements and cor
rective failures, the total replacement rate, Xt, is given by Eq. (9.22),
or
Ax = —Yi = Xp + Ac.
j0"R{r)dT
As defined in Section 9.5 the total average number of spares over a
long period of operation of the equipment, say t hours, is given by
EXAMPLE 14-7
Consider Example 9-3 and the system of Fig. 9.12. Determine
the total average number of spares for a period of 10,000 hr and the
number of spares at a 95% confidence level.
SOLUTIONS TO EXAMPLE 14-7
~Nsp = Ac t + Xp t,
= (A„ + Ac) t,
= (549 + 871)(10_6)(10,000),
or
Nsp = 20 spares.
where
(6 +0.5)- A' Sp
P(X < b) = $ (14.24)
7
Nsp
554 SPARES PROVISIONING
In this example
NSP
(NspY
CL = P(X < NSP) = £ * -N<
j=0
H0(T) = £-1Ho(s),
SPARES PROVISIONING 555
where
H0(s) = /(*)
Ml -/(*)]'
and
f(s) = C[f(T)] = Laplace transform of f(T)
For any given time period
t = j t; + r,
where
r t
j = /ivr (14.26)
LP J
is the total average number of preventive replacement cycles in the
time period, t, and
T = t-jTl (14.27)
is the remaining time after j preventive replacement cycles are com
pleted. The total average number of spares for a unit in the time period
t is the sum of the expected number of spares in j preventive replace
ment cycles which includes j H{T*) spares for corrective replacements
and j spares for preventive replacements, plus the expected number of
spares in remaining time r, H(t), or
where
H(T*) = expected number of spares needed in a time
interval (0,I£).
One additional spare is added due to a preventive replacement at the
end of each preventive replacement interval.
EXAMPLE 14-8
Given the unit of Example 11-1, determine the following:
1. The total average number of spares for a period of t = 10,000 hr.
2. The number of spares at a 95% confidence level.
556 SPARES PROVISIONING
10,000
j = INT = INT 33,
vJ 300
and
t = t-jr;,
= 10,000-(33)(300),
or
r = 100 hr.
Using Eq. (11.10) the expected number of replacements in a T*
interval is given by
4'
300 1 o 300
4- — e 200 — —
(2)(200) 4 4'
or
H{T1 = 300 hr) = 0.512 spares per T* = 300-hr interval.
Also
100 1 _2±92 1
H(t = 100 hr) = -I— e 200
(2)(200) 4 4'
or
#(100) = 0.092 spares.
The total average number of spares for t — 10,000 hr is given by
NSP = 33 [1 + 0.512] + 0.092,
77sp = 49.988,
or
N sp — 50 spares.
SPARES PROVISIONING 557
or
(NSp + 0.5) - 49.988
1.645 =
7.07
The value of Nsp which represents the number of total spares is
then found by solving for Nsp, and it yields
NSP = 61.12,
or
Nsp = 62 spares.
£?, V rrihJ
where
Ha(Tps), Hk(Tpf.) = expected number of failures in inter
vals (0,Tpj,) and (O.Tpfc) for s and k
groups of parts, respectively.
Cp, Cc = preventive replacement cost per unit
and corrective replacement cost per
unit, respectively.
558 SPARES PROVISIONING
The spares requirement model for any given operating time period t,
or the total expected number of spares for the ith part in Group s is
given by
WSP3 = Ks [l + Hi(T;s)} + H*(t),
» = 1,2, •••,5,
where
Ka = INT
pa J
Kk = INT
iT;k
and
rfc = t - Ku T*k.
The expected number of spares for each Group h part is given by
m — 1,2, h.
The actual number of spares for the tth unit in Group s, for example, at
the confidence level CL = 1 — q can be determined by using Eq. (14.23),
or
N'sp, _, (7v'p V
CL= Y,e-NsF'
1=0
(!
i = 1,2, — ,«.
Similarly, the equations for the actual number of spares for the units in
Group k and Group h can be obtained by using appropriate subscripts
and superscripts.
SPARES PROVISIONING 559
Ct = ^, (14.31)
-cy
where
and
EXAMPLE 14-9
Given the units of Example 10-2 with a Weibull times-to-failure
pdf, determine the expected number of spares, or the expected number
of spare orders to be made in a period t = 1, 000 hr of operation.
SOLUTION TO EXAMPLE 14-9
In Example 10-2 the total cost per unit time of operation given by
Eq. (14.31) is minimized and the optimum ordering schedule is found
to be T* — 23.5 hr. Using Eq. (14.32) the expected cycle length is
found to be Tcy = 29.463 hr. Then, the expected number of spares, or
the expected number of single-unit orders, in time period t = 1,000 hr
is
AT 1,000
SP INT
29.463
or
Nsp = 34 spares, or orders for spares.
where
is the expected number of less reliable spares for the units failing in
the interval (T0,Tp). E(r') is the expected residual life of a unit having
age T0 up to time Tp and is given by
!Jl R(T) dT
E(r') = (14.37)
R{TQ) '
Substituting Eq. (14.37) into Eq. (14.36) and adding 1 in anticipation
that at least one failure may occur in the interval (T0, Tp), the expected
number of the "less" reliable spares is given by
&:R(T)dT
NSp = A' T* + 1. (14.38)
R{T0)
EXAMPLE 14-10
Given the units with the gamma times-to-failure pdf of Example
10-3, determine the following:
1. The expected number of minimal repairs during t = 10, 000 hr of
operation.
2. The expected number of spares during t = 10, 000 hr of operation
and the total number of spares at a 95% confidence level.
A(T)
viv + T)'
562 SPARES PROVISIONING
Then,
or
l^^-Fwm*' (V + T)
E{Nno,T;]}=j-loge(l + -^
For tj = 200 hr the expected number of minimal repairs in a cycle
is given by
2. The expected residual life, E(t'), in a cycle for the gamma pdf
is given by Eq. (10.41), or
V _ p is
E[t'} = r; + 2 7?-e-^r-(r; + 2 7?)
v + T;
and the expected number of spares in a cycle is given by Eq. (10.42),
or
With X' = 0.05 fr/hr, the expected number of spares per cycle
becomes
200
E [N2[T;, t;]] = 0.05 j 210 - 185.5 185.5
200 + 185.5 L
+2 (200) - e-2,02oo55[210 + 2 (200)] }■
= 0.05 [24.5 - (0.5188)(585.5- 539.67)],
= 0.05(24.5-23.78),
or
EXAMPLE 14-11
Given the units with the gamma times- to-failure pdf of Example
11-3, determine the following:
1. The expected number of spares for the original units during t =
10, 000 hr of operation.
2. The total number of spares at a 95% confidence level.
3. The expected number of "less reliable" spares during t = 10, 000
hr of operation.
4. The total number of spares at a 95% confidence level.
1. In Example 11-3 the total expected cost per unit time of oper
ation function, given by Eq. (11.67), is minimized and the opti
mum values of T* and T* are found to be
T; = 300 hr
and
T: = 40 hr.
For the gamma pdf the ordinary renewal function is given by
Eq. (11.37), or
1 _23. 1
i .
4
With »7 = 200, hr it yields
40 r H 1 e ,jl 1,
H0(T; = 40 hr) = —; wo
2(200) 4 4'
or
H0(T* = 40 hr) = 0.01758 spares per cycle.
_.L_
566 SPARES PROVISIONING
l
j = INT
"10,000
= INT
300
or
j = 33.
Then,
r = t; + 60,
NSP = (j + i)h0(t;),
= (34)(0.01758),
or
n^QS / 1 + —77-
0.598 + -^7—
0.5982 .I = 0.9/ 1 .
0.95 < e-0598
1! 2!
Hence,
Nsp — 2 spares.
SPARES PROVISIONING 567
3. For the gamma pdf the mean number of "less reliable" spares
during the interval (T*, T*) may be found from Eq. (11.66), or
£[JV2[T0.,r„.]] = a jr; - t; j
e 1 (t; - t;)
+
0 -II (T.
b T'
+1
2 e 1 — + 1 ) -f p"' - - i
(t;-t;) 2
+ 2(2! -t;) + 2V}
t2>
e "—I + 1.
1 300
300J - (300)(40)
e '200
200
+200 [2 (300) -40 + 2 (200)]
(300-40)
(300 - 40)
+ 200 +1
200
«
2e - -i°- ( 40
20°l^5 \
+ 1J+e , jo
°' 200 — 1
(3°°-40>2+2(300-40) + 2(200)]
200
40
e 200—I + 1.
Then,
E[N2[t;,t;]] = 0002 I260 - 301.226
+0.136 [(200)(2.3)(1.635)
-(1,258)(0.03286)]}+1,
568 SPARES PROVISIONING
EXAMPLE 14-12
Given is a component with the gamma times-to-failure pdf of Ex
ample 12-1, with parameters
/3 = 2.5 and 7/ = 40 hr.
The cost of a corrective replacement is Cc = $100 and the cost of a
preventive replacement is Cp = $20. If the replacement opportunities
are exponentially distributed at the rate of u = 0.01 op/hr determine
the following:
1. The expected number of spares in t = 100 hr for the aptimum
replacement age, T", and the minimum total cost per unit time,
CT-
2. The total number of spares at a 95% confidence level.
3. Repeat Case 1 for the opportunity rate of u = 0.04 op/hr.
570 SPARES PROVISIONING
1. For given Cc = $100 and Cp = $20 the ratio Cp/Cc = 0.2. Also,
for u = 0.01 op/hr, /? = 2.5 and 77 = 40 hr the ratio, Cr, of the
mean time to failure to the mean interval between replacement
opportunities is
Then, from Table 12.1 for Cp/Cc = 0.2 and Cr = 1, the optimum
replacement age is T* = 0.25 m hr, where m = (i-rj = (2.5)(40) =
100 hr. Consequently, T* = 25 hr. Using Eq. (14.43) the mean
number of failures, before the preventive replacement age, T£, is
reached, is given by
N *-*TO
R(T;)= r f(T)dT,
„ = '-°fQ9Qf = 0.D63840.
0.939991
Using Eqs. (14.44) and (14.47) the mean time to the preventive
replacement age, L, is given by
JoT; R(t) dt
R(T;) '
Io5R(t)dt
0.939991 '
SPARES PROVISIONING 571
or
L = 26.092599 hr.
Using Eq. (14.45) the mean time the system remains above the
preventive replacement age, /, is given by
J~R(t + T;)e-»tdt
/ =
R{T;)
/0°° R(t + 25) e-°-01 « dt
0.939991
or
/ = 48.015763 hr.
Then, the average number of spares for an operating period of
/ = 100 hr is given by Eq. (14.46), or
NSp = (Ar+1),
L+l
100
(0.063840+1),
26.092599 + 48.015763
or
Nsp — 1.436 spares.
"sp (NspY
2 e~Nsp ., > 0.95,
;=o 3-
or
Cr = 4.
Then, for the cost ratio Cp/Cc = 0.2 and Cr = 4, from Table
12.1, the optimum replacement age is
T; = 0.37 m = (0.37) (100) = 37 hr.
Using Eq. (14.47) the reliability for the mission duration of T"
hours is
m 1 - 0-869486 m
0.869486
The mean time to the preventive replacement age, L, is given by
rS7
X = Jo jgg g = 40.717859 hr.
~^(37)
The mean time the system remains above the preventive replace
ment age, /, is given by
/0°° R(t + 37) e-004 ' dt
R(37)
J0°° R{t + 37) e-004 ' dt
0.869486
or
/ = 19.907018 hr.
Then, the average number of spares for an operating period of
t = 100 hr is given by Eq. (14.46), or
100 (0.150105+1),
40.717859+19.907018
or
jVsp = 1.897 spares.
SPARES PROVISIONING 573
"sr (NspY
J2 e~Nsp V ., ' > 0.95,
j=o J-
3=0 J-
Fig.
Component
14.1
MTBF
growth
pfor
Example
r14-13.
o-jection
Chours
oupmuelratitvieng
100,000
- 10,000
- 1,000
- 100
-J en
576 SPARES PROVISIONING
or
where
A,(Ta) = instantaneous failure rate function.
The average number of spares from time zero to the time of
175,000 accumulated hours of operation, when these components
reach maturity, or an MTBF of 4,100 hr, using Eqs. (14.51) and
(14.54), is given by
rl75,UUU
175,000 1J
NSp(0\ 175,000 hr) = j£ — T"0"3076 dTa,
175,000
1 j.(l-0.3076)
(1-0.3076)(100) a
= ^24 (175,000)(°-6924>,
__L_
578 SPARES PROVISIONING
or
NSp{0; 175, 000 hr) = 61.649.
According to the MTBF growth curve, the MTBF after 175,000
accumulated hours of operation should be 4,100 hr. So the av
erage number of spares during the time interval 175,000 hr to
1,411,200 hr, or until the end of the fifth calendar year is
FS,>(178,000 hr; 1.4U.M0 hr, = M11.200-17»,0l»
= 301.512 spares.
For the last two years, or from the end of the fifth year to the
end of the seventh year, the average number of spares is
or
JV5P(0; 33,600 hr) = 19.66 spares.
If we assume that the number of failures occurring in any time in
terval is statistically independent of the number of failures in any
interval which does not overlap the first interval, then the non-
homogeneous Poisson process may be applied. Using Eq. (14.5)
the upper, one-sided confidence limit on spares can be calculated
by choosing the smallest Nsp-Ui such that
P[N(t)<NSP-ui}= £ ± L ,
j=o J-
> CL = I -a. (14.56)
At the risk level of a = 5% the upper, one-sided confidence limit
for the first year is given by
Using the normal approximation to the Poisson pdf\ i.e., Eq. (14.24),
yields
(Nsp-ui + 0.5) - 19.66
$ = 0.95,
\/l9l36
or
(Nsp-ui + 0.5) - 19.66
= 1.645.
4.434
Solving for Nsp-ui yields
Nsp-ui — 26.454 spares.
Rounding up to the nearest integer yields
Nsp-ui = 27 spares.
For the two-sided confidence limits on the spares, round out to
the next higher integer value of Nsp-L2 such that
P[N(t)<NSP.L2]= £ 1 L ,
j=o
(14.57)
- 2'
580 SPARES PROVISIONING
and round out to the next higher integer value of Nsp-U2 such
that
NSp-v2 \NSP(t1;t2)Y e-Hspitito)
P[N(t)<NSP-u2}= £ * L ,
> 1 - |. (14.58)
Then, Nsp-L2 and Arsp-(/2 are the two-sided, lower and upper
confidence limits on the spares, respectively.
At the risk level of a = 5% the lower, two-sided confidence limit
on the spares for the first year is given by
*'•"•' * (19.66V _19.66 „ 0.05
El^-"u7 -19.66 <
i\ e - 2 '
,=o 3
or
Nsp-li
(19.66^ e-«u»
^ iif^ ,9.66 < 0>025
,-o i1
Using the normal approximation to the Poisson pdf yields
\(NSp-L2 - 0.5) - 19.661
$ = 0.025,
V19^66
01
or
Ns4=?>( 19.66V 1966
Y, r~ e-1966 > 0.975.
i=o •?•
or
Nsp-U2 = 100 spares.
According to the MTBF growth curve, the MTBF after 175,000
accumulated hours of operation should be at the constant value
of 4,100 hr. Consequently, the average number of spares for the
fourth and subsequent years can be calculated from
1
NSp(h\t2) = (t2~h).
MTBF
For example, the average number of spares for the fourth year is
given by
1,176,000-504,000
JVSp(504,000 hr; 1, 176,000 hr) =
4,100
or
7VSp(504, 000 hr; 1, 176,000 hr) = 163.9 spares.
The upper, one-sided and both the lower and upper, two-sided
confidence limits on the number of spares, at a given risk level,
can be calculated using Eqs. (14.56), (14.57) and (14.58), as il
lustrated for the first and third year.
The upper, one-sided confidence limits on the number of spares,
at various risk levels, are given in Table 14.4 and both lower
and upper, two-sided confidence limits, at various risk levels, are
given in Table 14.5.
Risk level a
Year Mean 40% 30% 20% 10% 5% 1% 0.5% 0.1%
1 19.660 21 22 23 25 27 30 31 33
2 40.226 42 44 46 48 51 55 57 60
3 81.962 84 87 90 94 97 103 105 110
4 163.900 167 171 175 180 185 194 197 203
5 57.366 59 61 64 67 70 75 77 81
6 229.463 233 237 242 249 254 265 268 276
7 245.850 250 254 259 266 272 282 286 294
Risk level a
Year Mean 40% 30% 20% 10% 5% 1% 0.5% 0.1%
1 19.660 17 16 15 13 12 9 8 6
23 24 25 27 28 31 32 34
2 40.266 36 35 33 31 29 25 23 20
46 47 48 51 53 57 58 61
3 81.962 75 74 71 68 65 60 58 53
90 91 94 97 100 105 107 112
4 163.900 154 152 148 144 140 132 129 123
175 177 180 185 189 197 200 206
5 57.366 52 51 49 46 44 39 37 33
64 65 67 70 72 77 79 82
6 229.463 218 215 211 206 201 191 188 181
242 245 249 254 259 268 272 279
7 245.850 234 231 227 221 216 206 203 195
259 262 266 272 277 286 290 297
SPARES PROVISIONING 585
The type of amplifier, failure rates, cost per unit and the number of
units in the assembly are given in Table 14.6.
The total time for which the spares will be provided is determined
in unit-hours of operation of the equipment over a three-year period,
for eight fire-control systems in eight submarines.
Three cases are considered:
1 1 r
Case 1 - 48,000 equipment hours - 3 years, 8 assemblies
(assumes 2.000 hr operation per year)
Case 2 - 9,600 equipment hours - 3 years, 8 assemblies
(assumes 100 hr operation per 90 day patrol)
Case 3 - 28,800 equipment hours, midway between that
of Cases 1 and 2-3 years, 8 assemblies
I0r-W
20,000 30,000 40,000 50.000 60.000 70.000 80,000 90.0C0 1C0.0OO
Total cost of kit. $
Jjs-c = 23.9,
NS-D = 8.16,
Ns-E = 8.45,
and
~NS-F = 4.80.
The optimum kit of amplifiers is determined using a form of the marginal
assurance as follows:
• _ logio Pna+i - logw Pna
P " Ci
or
P* = ^W10^ 1 (14.60)
x=0
where
xl
is the Poisson term, and N* is the number of a specific amplifier in the
optimum kit
A" = (7v;, n;, n;, n;, n;, n;).
SPARES PROVISIONING 589
6 *? _ (Ns-iY
t=l x=0
or
, E P2(*)
7T log10 -^ > 1.454 x 10'6.
E P2(X)
This yields TV^ = 41. Similarly the number of the remaining amplifiers
in the kit are determined yielding the optimum kit
.V" = (50,41,38,16,17,11),
590 SPARES PROVISIONING
£ w (NsA-iY
APA = £%-A^-,
e A__ L
t=0
N -0.
EXAMPLE 14-15
A system is to be placed in the field. The expensive, critical units
in this system and their characteristics are given in Table 14.7.
Do the following:
for Unit 2
JVs-2 = (0.000500)(6)(5,000)= 15 spares,
for Unit 3
]Vs_3 = (0.000750)(4)(5,000) = 15 spares,
The minimum cost spares kit for various confidence levels is found
by trial and error from the cumulative Poisson tables, or from
N,
P(*<iV,) = £p*(z) = E
N, -Ni
m
r=0 x=0
To find the confidence level of this optimum kit, use the val
ues in Table 14.8 of the cumulative Poisson probabilities corre
sponding to each N* in the optimum kit. In the optimum kit
JV*(36,22, 21, 6), for N{ = 36 the
N'
■-».„ KO'
X!
x=0
value is found to be
36 ,-24(24)a
Wat*)"E = 0.99178809
x=0
594 SPARES PROVISIONING
100
(40, 26, 2S, 7)
90
80
K 70
$
8
3c
Sa 60
o
O
* (32, 18, 17, 4
50
40 ■•
Fig. 14.3 - Optimum kits for minimum cost and desired con
fidence levels, P(N"), and N* are plotted versus
C{N*), for Example 14-15.
THEORETICAL CONSIDERATIONS 597
t=i
P(N') = (0.99178809)(0.96725576)
•(0.94689359)(0.98581269),
or
P(N9) = 0.8954798001.
The cost of the optimum kit is found from
C(Nm) = 22CiN?t
1=1
C{N") = (300)(36)+ 1,000 (22)
+(l,500)(21)+750(6),
or
C(N') = $68,800.
Figure 14.3 provides the plot of various optimum kits determined
using this procedure.
598 SPARES PROVISIONING
The first kit does not quite meet the goal. The second kit exceeds
the goal significantly; however, it might be possible to adjust the
units in this kit such that the goal is still met but at a lower kit
cost.
First, try decreasing 7V2 and iV3 by one (1), because they are the
costlier spares. Try
N* = (37,22,21,6); P(N') = 0.898374799; C(N') = $69, 100.
Since the 90% confidence level is still not met, try increasing the
least costly spare, N\, by one (1), then
N" = (38,22,21,6); P(N') = 0.900203220; C(N') = $69,400.
Since the 90% confidence level is met and at the lower cost of
$69,400, and other such adjustments do not yield the goal at a
lower cost than this, then the optimum kit is
JV* = (38,22,21,6).
Using the fill rate criteria the optimum spare parts kit N = (N\, • • • , Nk)
is determined so that the fill rate, Fr(N), is maximized subject to the
budget constraint
J2Nid<C0, (14.62)
»=i
where
k = number of different types of units to be spared,
d = cost of purchasing a spare unit of Type i,
and
C0 = available budget.
1=1
where
A, = demand or failure rate of a spare of Type i, i = 1, • • •, k,
ti = mean time to repair failed Unit i,
and
Ni = number of spares of Type i.
The algorithm computes a sequence of fill-rate-cost pairs which are
examined to determine a dominating sequence of allocations. The al
location ./V is said to dominate allocation N' if
k k
Fr(N) > Fn(N') while £ JV,- Q <J>? C,-, (14.64)
t=i i=i
k k
Fr(N) = FR(N') while J2Ni d <J2Ni Ci- (14.65)
t'=i t'=i
600 SPARES PROVISIONING
and
while
k
EN*ci <c0.
t'=l
EXAMPLE 14-16
A budget of C0 = $1,500, and the demand or failure rate, mean
time to repair and cost data are given in Table 14.11. Determine
the optimum spares kit containing these four units using the Kettelle
algorithm and maximizing the fill rate of the units.
SOLUTION TO EXAMPLE 14-16
In order to simplify computations use only the numerator of the fill
rate, Eq. (14.63), since the denominator is constant, or
1. Set up a table like Table 14.12 in which each row and column
intersection box contains a pair of numbers representing combi
nation of spares JVj for Unit 1 and JV2 for Unit 2, the total cost of
2
these spares, £ JV, C,-, and the corresponding fill rate numerator
for the given combination of spares.
2. Start filling in the table with (0,0) combination of spares which
gives zero cost and zero fill rate numerator. Proceed with cal
culating the cost and fill rate numerator for one additional unit of
each type. For example, consider combination of spares (N\,N2) =
(1,1). The total cost of spares is given by
2
£ Ni d = 1 (200) + 1 (100) = 300.
(*2 hY e-A2 «,
■a C •
■i—n J' ■i—n
or
Ni-l Ni-l
JV(1, 1) = 0.01 £ 022- e-1-0 + 0.02 £ £$- e"30,
jmO J- j=0 r-
= (0.01)(0.368)+(0.02)(0.05),
or
JV(1,1) = 0.00468.
602 THEORETICAL CONSIDERATIONS
Fill out the table completely keeping in mind that the entries are
made only if the cost constraint Ni C\ + Ni Ci < C0 is satisfied.
3. Determine the complete sequence of dominating allocations. Start
ing from the first combination of spares, (0,0), the next domi
nating allocation is the one with the higher fill rate numerator
but with the lowest cost. For example, the next combination is
2
(0, 1) with the total cost of £ N{ d = 100 and the fill rate
numerator of iV(0,l) = 0.001, or the next combination is (1, 0)
2
with the total cost of £ Wi Q = 200 and the fill rate numerator
of N(0, 1) = 0.00368. Since the combination (0, 1) has a lower
cost than the combination (1,0), then the second dominating al
location is (0, 1). To determine the third dominating allocation
consider the combination (1,1) with the total cost of 300 and the
fill rate numerator of 0.00468, and the combination (0, 2) with
the total cost of 200 and the fill rate numerator of 0.00398. Since
the combination (0, 2) has a lower cost than the combination
(1, 1), then the third dominating allocation is (0, 2). Continue
this procedure until the combination (iVi,JV2) with the highest
fill rate and the total cost slightly less or equal to the cost con
straint of C0 = $1,500, is found. If two or more combinations
with the same cost are found, choose one with the highest fill
rate. The complete sequence of dominating allocations for Units
1 and 2 is shown in Table 14.12 by the broken line connecting
the pairs of spares that satisfy the above conditions.
4. Repeat Steps 1 through 3 for Units 3 and 4. Table 14.13 gives
all possible combinations of spares for Units 3 and 4 satisfying
the budget constraint and the complete sequence of dominating
allocations.
5. To determine the complete sequence of dominating spares alloca
tions for all types of units simultaneously; i.e., Units 1, 2, 3 and
4, use combined complete sequences of dominating allocations for
Units 1 and 2 and Units 3 and 4, respectively, and the body of
the table is filled out so that for a given combination of all units,
(Ni,N2,N3,N4), the total cost and the fill rate numerators are
summed up. The complete sequence of dominating allocations is
obtained by using Step 3. The broken line in Table 14.14 rep
resents the resulting sequence of dominating spares allocations
which satisfy the budget constraint of C0 = $1,500. The last
combination in this sequence,
{NUN*N*NA) = (0,6,3,0),
604 THEORETICAL CONSIDERATIONS
TABLE
aP14.14
of
kits
lwith
otheir
aUnits
for
socosts
2,
1,
cspares
oai-ctbiloantesd straight
the
in
line
body
iathe
table
ofdleonmtciantfiaoetnsi.ng
0.02862 3.7,0,0 0.02862
(3.7) 1,300 1,300
(*.«)
0.02366 2.5,0,0 0.02366 1,6,0,1
^2,5,0,1 1,050^ 0.02501 2,5,1,0
0.02335 0.02861 2,5,1,1 0.02096 2,5,2,0 0.03755
(2.5) 900 900, 1,150 1,200 1,460 1,500
Unit
Type.
0.01998 1,6,0,0 0.01998 1,5.0,1 0.02133 1,5,1,1 1,250 0.02628 1.6,2,0 1,300 0.03387
1.2 700 700 960
0.01832 0,6,0,0 0.01832 0.6,0,1 0.01967 0,6,1,1 -1.150 0.03462 0,6,2,0 1.200, 0.03221 0,6,2,1 0.03356 0,6.3,0 0.04025
600 850 1.500,
(0.«) •00.
S1,450
0.01630 0,5,0,0 0.01630* 0,6,0,1 0.01765 0,5,1,0 0.2125 0,5,1,1 1,050 0.02260
- 0,6,2,0 0.03019 0,5,2,1 0.03154 0,5,3,0 0.03823
(0.5) 500 600 760 800 -^1,100 1,350 '1.400
0.01294 0,4,0,0 0.01294* 0,4,0,1 0.01429 0,4,1,0 0.01789 0,4,1,1 0.01924 0,4,2,0 0.02683 0,4,2,1 0,4,3.0 1,300^0.03487
(0.4) 400 400 650 700 950 1.000 1,250
0.02S18
0.00846 0,3,0,0 0.00846 0,3,0,1 0.00981 0,3,1,0 0.01341 0,3,1,1 0.01476 0,3,2,0 0.02235 0,3,2,1 0.02370 0,3,3,0 0.03039 0,3,3,1 0.03174 0,3,4,0 0.03519
(0.3) 300 300 550 1,150 1,200 1,450 1,500
600 850 900
0.00398 0,2,0,0 0.0039S 0.2,0,1 0.00535 0,2,1,0 0.00893 0,2,1,1 0.01028 0,2,2,0 0.01787 0,2,2,1 0.01922 0,2,3,0 0.02591 0,2,3,1 0.02726 0,2,4,0 0.03071
(O.J) 200 200 450 600 750 1,050 1,100 1,350 1,400
800
0.00100 0,1,0,0 0.00100 0.00235 0,1,1,0 0.00595 0,1,1.1 0.00730 0,1.2,0 0.01489 0,1,2,1 0.01624 0,1.3.0 0.02293 0,1,3,1 0.02428 '0,1,4,0 0.02773
(0.1) 100 100 350 400 1,000 1,260 1,300
660 700 950
JB.1,0,1
Unit
Type* numerator
0.00135 00435
0 0.02808 0.02892
3,
4 Sparei Fill
rate (0.0) (0,1) 250 (1.0) 300
(Ml 550 0.00630
(J.o) 600
0.01389
(2.0 850
0.01524
(3.0) 900
0.02193
(3.0 1,160
0.02328
(4.0) 1,200
0.02673
(4.0 1.45U (s.o) 1,500
0 0
Co«l
606 SPARES PROVISIONING
ftOT.""*
v ; 0.07 0.460.
The optimal spares kit and fill rate obtained for a budget constraint
ranging from 0 to $1,500 is given in Fig. 14.4.
If there are more than four (4) types of Units, use the same proce
dure until all types of units are combined and the optimum spares kit
(Ni, ■ • • , Nk) is determined. Apparently, as the number of unit types
increases the number of combinations or allocations that satisfy the
budget constraint increases. In that case computer optimization and
search methods should be used.
The major drawback of this approach is the assumption that all
failures are of the same criticality which is applicable to a single system
without redundancies. In the case of a system with redundancies and
multiple operating modes this assumption is not valid. The UNIRAM
availability model [9] enables the selection of the most critical parts,
units or subsystems by ranking equipment, parts or units based on
their outage rates or unavailability. The model is developed through
several steps as follows:
THEORETICAL CONSIDERATIONS 607
0.6 - (0,6,3,0)
(0.5.3,0)
(0,4.3,0)
0.5 - (0.6.2.0)
a (0,5,107
2 0.4 - (0,4,2,0)
— (2.5,0,0)
ii. g,6.6ToT
0.3 "
(0.5,0,0)
0.2 - (0,4.0,0)
(0,3,0.0)
0.1 - (0,2 .0.0)
(0,1.0(0) 1,500 -v
(0.0
i i i1 1t 11 4-1
It 1 1 1 1 1 1 1 l
0 200 400 600 8oo ' l.ooo'iioo'i,^'
Cost in dollars
Fig. 14.4 - Optimal spares part kit and the best fill
rate obtained for the various budget con
straints of Example 14-16.
608 SPARES PROVISIONING
2. Develop the fault tree model of the system and model the failure
of its subsystems. All the failures that cause unavailability of the
system are modeled down to the component level or basic events
of the fault tree.
A MTTR.+MTBF,
A-i MTTR, A "J.
f-> /i|
t=l _ i=l
MTTR, ~ -J. '
MTTR.+MTBF, "*
where
FR(N)=i=1 >=\_ .
1=1
where
Pi = P[N(t) = i] = probability of demand of i units.
Similarly, a shortage of spares occurs when the actual demand of spares
is higher than the stock level. Then, the expected number of shortages
is given by
X) («-*)»■ (14.70)
i=x+l
The total inventory cost is the sum of the expected overstocking plus
expected cost of understocking, or
where
Ch = holding cost per one unit,
C'sh — shortage cost per one unit.
610 SPARES PROVISIONING
The optimum stock level which minimizes the total expected cost is
given by
P[N(t) <x0-l}< °s" < P[N(t) < x0}. ( 14.72)
EXAMPLE 14-17
A repair crew is to obtain enough spares for 3 equal units, having a
failure rate of A =1/6,000 fr/hr, for a field repair mission which should
last 4,000 hr. If the cost of carrying a spare is Ch — $1,000 and the
penalty cost of not having a spare when needed is Csh — $10,000,
determine the optimum number of spares in the kit.
SOLUTION TO EXAMPLE 14-17
Assuming a constant failure rate, A, the number of failures, or the
demand in the time period of / = 4,000 hr for 3 units, is Poisson
distributed, with the expected number of spares, or expected demand,
given by
3
Nsp = X>>,,
1=1
3
= Aj>,
1=1
P[N(t) = i] = ^-^e-N^,
To determine the value of the optimum spares kit, x0, create a table,
with the cumulative Poisson probabilities of Eq. (14.73), as given in
Table 14.15. From Table 14.15 the value of the optimum spares kit
which satisfies the inequality of Eq. (14.74) is x0 = 4, or
_SP_e-NSP
E^f'"*" < 0.909 < J2
:=0 i=0
or
P(A) = f[Pt(xoi),
or
m
P(A) = J[P[N(t)<xoi), (14.75)
»=i
where
P(A) = confidence level, or probability, of not running out of
spares for the whole system,
and
Pifeoi) — confidence level, or probability, of not running out of
spares for the ith unit.
Difficulties that arise in the application of this model are the determi
nation of the penalty costs due to a shortage of spares for all units in
a system. Assuming that the penalty cost is equal for all units, and is
related to the cost of aborting the mission, application of this model
is limited to one shot items or mostly to weapon systems.
PrW, = ^ + 2f (14.76)
where
D — average demand of identical units over the total pro
visioning time,
Ch = holding or carrying cost of a unit per ordering inter
val,
C0 = order cost of a quantity Q, or
Co — Cs + Cu Lj, (14.77)
where
Cs = setup cost per order ,
and
Cu = cost of a unit.
Substitution of Eq. (14.77) into Eq. (14.76) yields
EXAMPLE 14-18
An inventory of spares for Traveling Wave Tubes (TWT) is to be
established. The expected tube usage over a 2-year period is D = 500
units and the holding cost of a unit during the stocking period is C/> =
$300. The setup cost of each order is Cs = $800, the cost per unit
is Cu = $50 and the procurement lead time is 2 months. Determine
the optimum ordering quantity, Q", reorder point, r, and the ordering
cost, C0-
SOLUTION TO EXAMPLE 14-18
Using Eq. (14.78) the optimum ordering size is
Then, from Eq. (14.77) the ordering cost corresponding to the opti
mum ordering size is
Co = ^s + Cu Q,
= 800 + (50)(51.64),
or
Co = $3,382 per order.
Since the average demand of spares, D, is given for the period of 2
years, then the average demand of spares during the lead time of 2
months is given by
_S_ 500
2x12 24 '
or
2 = 41.67 spares.
Rounding up to the nearest integer yields
z = 42 spares.
THEORETICAL CONSIDERATIONS 615
or
3 V^ = 20 spares.
r = z + 3 %/l,
= 41.67+19.36 = 61.03.
or
r = 62 spares.
Hence, when the stock level reaches the level of 62 units an order of size
181 units is made. The reorder level of 62 units assures with 99.9%
probability that there will be enough spares to supply the demand
during the procurement lead time of 2 months. Figure 14.5 shows a
typical stock of spares flow, with the stock parameters Q and r that
were calculated previously.
62
level,
Reorder
r=
20 Example
14-18.
flow
for
of
stock
Typical
Fig.
14.5
spares-
6
18
16
112
8
14
4
20
(
\
Re
\
Lead \
itime
i\\
i\i \
i\i \
1 H
(——
1
months
T,
Time,
\
Lead \
timei
1
Hr
i\ \
i\i\
i \1
i
o a ■o a. i o0
b £
en >— Oi
THEORETICAL CONSIDERATIONS 617
Time, T, years
or setup cost, the expected holding cost and the expected shortage
cost, and is given by [12, pp. 54-93]
CT(Q,r) = C D + C0^
where
C = spare part purchasing cost,
D = average demand, units/year, or
D = 7i A (365), (14.81)
C0 — cost per order or setup cost,
Ch = holding cost per unit per year,
Q — order quantity per cycle,
and
r = inventory reorder level.
The expected demand, E(x), is given by
f°°
E(x) = / x f(x) dx, (14.82)
Jo
where
f(x) = pdf of demand, X, during lead time, t.
and
l(t) = pdf of lead time, t.
Assuming that the lead time is lognormally distributed, then
1
l(t)
t a/2 7T 0ff
r*^)1 (14.85)
where
t' = loge t for 0 < t < oo.
The mean and standard deviation of the variable /' are given, respec
tively, by
and
ot< = (14.87)
where /z* and o^ are the mean and standard deviation of the straight
times, t. Since all n units are identical with constant failure rate, A,
then the overall arrival rate is (n A) and the conditional pdf of demand,
x, is given by
Since g(x,t) is a discrete pdf then the pdf of demand, x, given by Eq.
(14.84) can be written as
The expected demand during lead time, E(x), given by Eq. (14.82),
can be written as
L._ _
620 SPARES PROVISIONING
and the expected shortage during lead time, S(r), given by Eq. (14.83),
may be written as
oo
S(r) = £(* - r) p(x). (14.91)
X=T
*cv = |- (14-95)
Since the values of k and m are real, round them up to the nearest
integers. The modified cost function is given by
mean value of the procurement lead time is 90 days with the standard
deviation of 45 days. Determine the optimum ordering quantity, Q,
and the inventory reorder level, r, so that the total expected modified
cost is minimized.
SOLUTION TO EXAMPLE 14-19
To determine the modified cost function, CPA(Q,r), the first step
is to calculate the discrete pdf of demand, p(x). For given values of
Ht = 90 days,
and
<rt = 45 days,
the mean and standard deviation of the lognormal pdf of the lead time
are obtained from Eqs. (14.86) and (14.87), or
90
IH> = l°ge
f*W'\
= loge[80.498],
or
Hf = 4.388,
and
Of
i
loge 1 +
(sy
= \/0.223,
or
at. = 0.472.
Then, the pdf of the demand, p{x), during the lead time, given by Eq.
(14.89), is
v(x) = ___i r [(20X0-0025) ty m (o.oo25) t]e^c^j^if
yy ' JT^ (0.472) Jo x! * '
or
Since Eq. (14.97) can not be integrated directly, then Eq. (14.99)
must be solved numerically; i.e., for each value of x the integral given
by Eq. (14.97) should be determined by using, for example, Simpson's
rule with the sufficient number of intervals.
Using Eqs. (14.80), (14.91), (14.98) and (14.99) the total expected
inventory cost is given by
1 o or
Q
CT(Q, r) = (65) (18.25) + (23)-^ + 50 J + r - 4.866
(66)(18.25)^2», > , >
+^75 £(* - r) P(*).
or
419 15
CT(Q,r) = 1, 186.25 + ^p + 50 5 + r - 4.866
A(Q,r) = ±{-^le^"^,
i=0
or
A(Q,r) = f^[k (°-00f } tcy3'e-^ (°-0025> ««J. (14.101)
i=0
The total expected modified cost is given by the ratio of Eqs. (14.100)
and (14.101). Since there is no analytical solution to this optimiza
tion problem, numerical optimization should be used. A simple search
technique may be applied using the following algorithm:
THEORETICAL CONSIDERATIONS 623
m = Q + [r - E{x)},
= 8 + [5 -4.866],
or
m = 9 units. (14.103)
Then, from Eqs. (14.101), (14.102) and (14.103) the maximum proba
bility of having a spare when needed during a cycle is given by
4(8,5) = V* - (Q-0025) <c*]' c-f* (0.0025) tcy)
t=0
9
= £ [(20) (0.0025),'! (159.87)]' [(20) (00025) (15987)]
i=0
(0.000338) ^(7-",35)'
i=i l-
or
4(8,5) = 0.7182. (14.104)
Hence, for given minimum total cost, the maximum probability of not
running out of spares in a cycle is 71.82 %.
"0 (O mm ttSo>«r«t(oa)fiNticBOr<t(P(OOr<<reoOf<niAM])Nn>nsaNt
Ob r- p- r- r- t.NK(ooo«no»ftftO)SiJ)8oooo»"""H*,"WM«Nr)nnnntt
c l« W « W « I Ci P* C* ' (N « N M I
a
oo Co oi^ n»i-h-»or)«-iOQaa«OQP«-» — c*rtcovio«r-«aQ«wn -!■ iO CD p- 01
ic„J«wr)ioiotop«*iDsS»«^a«)pNt»floortT«)BonioNO) — CO m P- 01
t- r- r- r- m> sr-i-ctCBiuiBttOiftftoOOOO- — — — " r< m ts n n co co co co co
iN««MC*r»r«wp<oip«c*ciwe»»
TiOWNOOOlO'-nfWtt
p. «■ t- ms toAONtiSNOiH(>)iAhO)*ninNOi^ftpooor«v<Q(CONiosa*<nia
c
cti
3 — I- 1* p- NNnonMfiioiAvtieiAinettNtatoO) Ox n n
cr ▼ r* n c« S
X C (C c
01
CO
tO
micB)o«n»Ncjif»nioNsi»-niof»ftH ▼ 5 ee o
ClilCNt-t-r-SSflOOOOOOOWOlClClOftO o 5 o —
CN ■v to
SB
c c
-p p*
0) O
n it.
p* p*
"
i-
p«
nn
a -
c* co
k
r- 0) p- p- Oi 00 oint-QOfflmYntiNdNnnnvmiofflt- o* ro <r to (0 n OS o ••
-o q co o> 00 OO
m
a, o> .- co §
& o o 5
00 t V B O N MS
C 00 CO IO lO to - ~ « n N <N N
iO*WO*C«P*0«IP«NNP«OI
O
i naiBiOBO ▼ l. .... _ ... _ . ._.-._. „.. _-
8 m vW iASBO"nmKoi<"niAhoit<nnNaif<ninKoiMp:
to «o to totoiotototototOgop-r~i^p-F-oooooo«)oocfl0i0)C>io>O©i
i r* c* c* p* p* r* r* <
11 ^
to o n O'
ifl r» n »• . . . . . . iMTBO)*P)lflhO>HniOt»Bi"n(PBOMT(C«
O tO tO tOltO tOtOtOtOlOkOC0tO NhSKBBBBBQOiOlOOlOOOOn'-xx-
54 iW(i«NNNNM<S
P* lO CO ■ I Soiaio-'-fionvifliDSBffio-N
r* T P- < Nvsffl"ninsoi«nwKOiN*c
NNNhhVIBB'KaOOIOlOlftOlOOO
C 0 ■ P« P« P*
.2
*» J « co -w — 01 Qoli * B o MOOS io n n « - O O —
o _r PJ fi O O » 00101 O *- CO t » r- o> —
to
CO
to
m
to
P-
to
01
to s I to
tO
p-
to
01
tO
p-cotor-o>^^cotor-oj^-cotooo Q P*
Nt»NI>-f-BBBB00OlCftO)Ol C O
c v P* P»
3*
<*• >
Q) BWMiANHnOpn to p- oo ^r n o n s <c to
n >n v O) V p-Ip- oo 0> O «- CO ^ to oo © >~ co to p-
to
01
■■? *r ^
— CO tO
T*»«WfflONSBBftO»- P« P*
soi-nwt-ft-. n w n ft w v to X
onfl^n n colco co co ^ ^ "T ^ f ^ to to to to »o lO to to to BBfSNSNBBBBBOlO) 0> a
pp P"" P- 4Pt P« — 1— PP PP PP> M — M AM ■a pa p« m « M V pp pa pp M«N*«*«««*Mpa<w*MpM««««MM ■m pa
8
0 8
D
o "0 Pt o CO p- to CO p* O O O O o
3 Pi ro ro
p* *■
-o o
h — p*
*- to
©
9
r-
ro
to
CO »* p- oo
n b
01 pt co to p- 8 8m p*
to to
to eo
tO tO i P*
to to 1 S8 to p-
p« 00
p.
o
CO or CC
to CO
00 S 01 01
s-a >> ■
(- p*
oo n n 01
CO p* olco 01
88 8 S
00
8eo p*
o
CO p*
CO
p-
to So
00
01
to
CO
eo
to
P4
P- so 01
p*
00 00
i
CC cc
O
F
p-
p«
OO
to oc
OD CO
s 00
p*
cr. 01
to S2
iE £s CO r-
D CO
lO
CO
CO
to
o
lO
CO
00
CO
n
o 00
CO
o
co
P-|0>
88
CO
tO
p-
CO
CO
8
CO
0)
CO
o
r~
CO
to P*
to
00
to
tO
0.
tO
p*
to
tO
01
p*
tO
»n m to to to
p-
to
S
to
oo
m
ro
S
P<
to to
to
s
3
a
p»
CO
1
to
00
p-
00
co
p-
ii-i
CO
00
00
p-
01
CO
CO
00
X
CO
CO
CO
oo a
oo ac
tO
oo ac
P-
H .S oi p- CO to O tO
01 00 T p*
a
co ■p co to 8 ro 01 ■f
to
P* 0C p-
co
iO P4 o c
c o CO
I
01
p«
to
— s tO CO
OO
co rt
00 00
CO % 01
CO
p* to 00
to m
OO
§ to tO
to
to to
CO m P-
P.-P-
?. tn
CO GO oo
to a pt •o s- 01 r-
oc PS
■ -r
CN
i lO
CO
lO
00
■J
p-
■J"
w
to ^,
P4 PP
p.
o o
p* oo p- p-
CO
•r 3 00
CO
to
o
m
00
to
CO
lO
O
in
m
t~
to
iO
00
in
s
CO
0) p-
ro m
CO
CO
p- 0)
te cc
p* O 0)
t-
00
r-
co ac 85
CO CO
W ■ CO
— r- CO CO
CO
to
O
00
00
to
"W
^^ in
CO CO CO CO
01
CO JJ
m
in to 00 i cc
p*
ro CO
m a cr
Q
CO
P4
O cc m ro
m p- 01
o
CO
cc
03
tO
CO
V
o
<c oc 01 to it: If w tn m m m tO |Q to tc CO to tc tc P" p> r» cc
■T C-* P*
PD
< 12: p*co^rtntop-oo0iO
624
THEORETICAL CONSIDERATIONS 625
losses. Spare parts of low criticality are not significant either to mis
sion success or to financial losses. If out of stock and needed they can
be substituted by alternate parts or they can be found in the market.
Selection of the optimized inventory policy depends on the accurate
assessment of criticality. Once the cost of purchasing and ordering,
as well as the stock-out cost, are analyzed a decision on a spare parts
provisioning policy can be made. The following methodologies based
on parts classification could be one possible way of decision making:
Strategy 1
If purchase and order costs are high and the criticality is varying;
i.e., the parts are classified in Class I in Table 14.17, keep the safety
stock level of spares which are to compensate lead, repair and shipment
times. Place an order on a one for one basis whenever a failure occurs.
The initial stock level can be calculated by using Eq. (14.5), or
— \k
Ns
E .-*,H k\
= CL,
k=0
which would give a sufficient quantity of spares with the desired confi
dence of having enough spares during lead, shipment and repair times.
Alternatively, Eq. (14.79) may be used if the normal approximation
to the Poisson pdf is used, or
r = z + 3 y/l.
The average demand, z, is taken to be equal to the average number of
spares, N s, and 3 V^ assures 0.999 probability of having enough spares
during the lead time. The average demand, z, may be also estimated
from the field data or past experience.
Strategy 2
If purchase and order costs are medium and criticality is varying, or
the purchase cost is high with low criticality; i.e., the parts are classified
in Class II, inventory costs such as ordering, holding and shortage
PROBLEMS 627
costs should be considered and the appropriate model for this strategy
selected from among the inventory models presented previously.
Strategy 3
This strategy deals with items of low cost and varying criticality
grouped in Class III. If there is no degradation of characteristics, stor
age cost is low and the parts are not instantaneously available in the
market, this type of parts should be ordered in one large quantity or
for the whole mission duration or for the system's life cycle. Ordered
quantity of size, N, should be determined by Eq. (14.5) so that the
desired probability of not running out of spares during the mission,
or the desired confidence level is achieved. If some of the conditions
mentioned above are not satisfied then consider an inventory of spares
model such as in Strategy 2. If the parts are available in the market
when needed, then maintain a stock at a local supplier or use just-in-
time techniques.
It is obvious that the optimum spares provisioning is a very com
plex problem and it requires accurate analysis of all conditions and
factors that may affect the selection of the appropriate sparing model.
Costs of ordering, holding, and understocking should be estimated,
failure rates predicted, criticality analyses performed and preventive
maintenance policies considered. Incorporating all of these elements,
in a well planned methodology, will result in an optimum spares provi
sioning policy with reduced total cost and a maximum confidence level
of having a spare part when needed.
PROBLEMS
14-1. Spares need to be provided for a critical unit in a system for 500
cumulative hours of operation at a confidence level of 95%. If
the unit's failure rate is 0.009 fr/hr, how many spares should be
provided for this period of operation?
14-2. Work out Problem 14-1 again but at confidence levels of 80%,
90% and 99%, and discuss your results comparatively.
14-4. Work out Problem 14-3 at the confidence levels of 80%, 90% and
99%, and discuss your results comparatively.
628 SPARES PROVISIONING
(3) Tp = 00 hr.
14-7. A system consists of two equal units in parallel. Each unit has
a failure rate of 0.001 fr/hr. Determine the number of spare
units required for a 3,000-hr period, if the units found to have
failed during the maintenance action are replaced and all units
are kept in their useful life. Consider the following preventive
maintenance schedule times:
(1) Tv = 10 hr,
(2) Tp = 150 hr,
and
(3) Tv = 00 hr.
PROBLEMS 629
J. _
632 SPARES PROVISIONING
REFERENCES
635
636 DISTRIBUTIONAL TEST AND REPAIR TIMES
At station No. 1
t, = 11.3hr.
7 8 9 10 iTf 12 13
At station No. 2
f,(t) o-2- t, « 17.5 hr.
. t
16 17 \ 18 19 20 21 22 23 24
At station No. 3
4= 1.9 hr.
(5 I p 3
At station No. 4
t. = 6.3 hr.
0 2 f 8 10 12 14 16 18 20
At station No. 5
U = 0.3hr.
At station No. 6
t« = 9.2hr.
5 E(\:= 5)
N = normal pdf,
G = gamma pdf ,
E = exponential pdf,
X2 = chi-square pdf,
fi = mean,
a = standard deviation,
/J = shape parameter,
7/ = scale parameter,
v = degrees of freedom.
638 DISTRIBUTIONAL TEST AND REPAIR TIMES
and
P{t < 65.2 hr) = 99%,
and
d2h d2h . . _
—j means —j evaluated at the mean value of all
' ' variables involved, or of the t's.
Most books only give the first term of Eq. (15.3) and consider the
accuracy sufficient. The first term only provides an exact result for
our problem because all second and higher order partial derivatives
are zero. Equation (15.3) contains only up to second order terms of a
multivariate infinite Taylor's series. Hence, it is only approximate in
the general case. Also,
ti2(z) = VAR(z) = ol = ti-(ti)2,
where
H2{z) = second moment about the mean.
The Taylor's series expansion for the variance of the system's perfor
mance, with uncorrelated variables and retaining up to third order
terms, is given by
where ^3{ti) is the third central moment for the ith variate. Most texts
use only the first term, which may be a satisfactory approximation.
The third central moment is given by
retaining only the lowest-order non-zero term, and the fourth central
moment is given by
vr\ 4
M*) = E 1=1
I) "<"•>
n n— 1
dh dh
VAR(ij) VAR(U) . (15.6)
at. du
HOW TO APPLY 641
(15.8)
N
IN \
E*3 E«? E* 'eO
Mt) = JV _ 3
.=1 i=i_ , i=±_ + 2 (15.9)
tf N
and
Also,
m<«) = Mi - 4 M3 Mi + 6 Ma (m'i)2 - 3 04 )4 . (15.13)
It is related to the coefficient of kurtosis, a4, which is given by
<*4 = P2 = §. (15.14)
M2
In general
4(0 = £(«*), !
-jT«*/W*i for A = 1,2,3,4,... (15.15)
= fcth raw moment of the distribution /(t),
= fcth moment about zero, or about the origin,
642 DISTRIBUTIONAL TEST AND REPAIR TIMES
and
1 =
2>
t=i
Then,
Station number
1 2 3 4 5 6
Mean 10 20 1.50 10 0.20 10
Variance 1 2 0.25 10 0.04 20
0 0 0.08 20 0.02 80
3 12 0.23 360 0.01 1,680
Then,
t = 51.7 hr; a\ = /x2 = 33.3 hH = ft^x); at = 5.78 hr;
1. Find
1.1 -1 = 1.
1.2 - a2 = at = y/p,2(z)-
1.3 - $i = &l
1.4 - fa = Q4-
2. Find in Tables 15.2 thru 15.15 the standardized percentile z'a for
a chosen a using /?j and fa.
3. Calculate the estimated 100 a% percentile from
or
ta = t + za at,
and
P (t < to.90) = 0.90.
In this case
and
P (t < to.99) = 0.99,
where
^n — zn+l ~ 2 2„ + zn-l> ^n+1 — zn+2 - 2 Z„+l + zni
and
* = [Pi-Pun)] 10-
The intervals of Pi used in the interpolation of ft must be equal;
i.e.,
EXAMPLE 15-1
Given £ = 3.49 and ft = 0.27, find the upper 10% percentage
point, a = 0.90, z' value.
/$,
expressed
given
for
and
of
62.
points
PBased
"Table
pKrivoy,
•Prepared
by
Raul
eMr.
conarucrsevonetsa,ge
1158 1366 1488 1554 1585 1593 1586 1574 1552 1535 1508 1480 1453 1433 1406 1386 1366 1339 1323 PiNixon
S.
E.
by
snwith
Compiled
Eric
and
JetN.
L.
I).
Amos,
min
roaeandrhnuscdtuosiarnoe,nd,"
).70
1
0.60 1 1.1625
1.1501 1 1.1687 1 1.1530
1 1 1 1.1446
1 1.1424
1 1 1.1373 1
1.1404 1 1.1353
1 1.1710
1 1.1710
1 1.1695
1 1.1675
1 1.1649
1 1
1.1616 1.1593
1 1.1560 1.1499 1.1477
1.1278
u)/a;
10%
of
dspoints
(x
0.10)
the
(a=
Lower
signs
All
tneanvdgiar-adtizevde*
1655 1.1781 1.1837 1.1851 1.1840 1.1816 1.
1. 1. 1.1437 1.1407 1.1387
782 1.1753 1.1714 1.1675 1.1645 1.1606 1.1576 1.1546 1.1516 1.1486 1457
1
0.50
1.1839 1.1965 1.2012 1.2012 1.2007 1.1951 1.1907 1.1861 1.1817 1.1777 1.1729 1.1692 1.1652 1.1615 1.1586 1.1555 1525
1. 1.1494 1.1468 1.1444 1.1418
0.40
1. 1.2183 1.2214 1.2197 1.2155 1.2102 1.2046 1.1990 1.1935 1.1878 1.1834 1.1783 1.1737 1.1697 1.1656 1.1621 1.1592 1.1562 1.1532 1.1501 1.1482 1.1451
2059
0.30
POINTS
PCURVES
OF
ERACERNSTAOGNE
1.2342 1.2454 1.2462 1.2419 1.2354 1.2280 1.2208 1.2135 1. 1. 1.1631 1601
2069 1.2003 1.1943 1.1887 1.1837 1.1786 1.1747 1.1707 1666 1. 1.1571 1.1541 1.1516 1.1491
0.20
1.2394 1.2589 1.2625 1.2585 1.2512 1.2427 1.2341 1.2259 1.2179 1.2108 1.2038 1.1976 1.1916 1.1866 1.1817 1.177S 1.1735 1.1695 1.1655 1.1625 1.1595 1.1565 1.1535 1.1515
0.15
TABLE
15.2
1. 1.1764 1.1724 1.1684 1.1654 1.1624 1.1594 1563
1. 1.2505 1.2406 1.2309 1.2229 1.2151 1.2080 1.2012 1.1952 1.1902 1.1852 1804
1.2747 1.2828 1.2794 1.2710 2609 1. 1544
1.
0.10
0.05
1.3104 1.3074 1.2972 1.
2846 1.2818 1.2596 1.2481 1.2381 1.2293 1.2205 1.2136 1.2066 1998
1. 1.1948 1. 800 1.1760 1.1720 1.1691 1.1661 1.1631 1.1601 1.1573
1890 1.1848 1.
1
1963.
aBVol.
44,
3
Nos.
50,
i5opp.
9ndm-et4r9i8ka,
0.03 1.3189 1.326S 1.3188 1.3057 1.2913 1.2773 1.2643 1.2524 1.2421 1.2329 1.2238 1.2166 1.2093 1.2025 1.1973 1.1915 1.1873 1.1823 1.1783 1.1743 1.1710 1.1680 1.1651 1.1621 1.1590
1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8 S.O 5.4 5.6 5.8 6.0 6.2 6.4 6.6
\
B2 S.2
o
00
OVOOOOOOOOO
I?
c x:
co^-r^r^ioomao^fO 11 *j
-Hro^mvor*.i^.r^Qooo > •««
m oooooooooo -H 3
an M
o
c H0
i/l
0
IM f=
g, o
oovOfM^r^or^of^o»-tt^
OWhwnnnuimjio
<
0<M^i/>vOP»r»ooooooooo t/l to.
oooooooooooo 0 IU ifl
>(- m
-H
3 a
U ■a ■
CO
c c en
o o to T
!N"f>0NMO00!Jl<JlOiaiOl in 1
oooooooooooo fc. c
to a
C71
in
0) X T
0.
to 7
IM a.
O C
06 O
OMm\or~ooa>oiwooooo VI u, *
0O0O00O0O"--"--«-<>-« *->u *T
B
•-* ■»■a
o c c
c o to
^*r^«©K}^*r4co<NOis-r^~*»0\0 V c ei
^•■Or^aOOiOOOOOOOO© oo
CO
j:
o •
oa©o©~»<-«**'-«-<'^-«'-* *j
c
'-> i/i
o
<u
o —j
^;
E ■
u a
w^Nso^Hrjini^M^siOi/iinui cz in
r^r-at^'to>^oo--M^'inm^*tf(OfN
OOOOO-***-* — -* — -Hrt«-*rt
<M
o XI
>-, •
o
~* ^■ >■
V •o
X>
to •** to
f- C*
ror"*Ki*TCNr*»o*-*CNc^(N«oo*coi"* ^ B —<u
o
000«-H-»-«-<«-H-H — «,„„,< c u w
o
-a ■ 10
a a •^
in u C3
<o 3
o ^j-oirtr^^y^HsoooooocAoor-^^-vKifsi CO 1/1 >
o to B
OOOrt-,*-<-N-H«^-..H — -*.Hi-l-<>-« » <a O
>,
0
1 fc.
in
■tH> -nt- ■a
^M -atoc c
o ~H to t/i
o 3 *i
■
c o ro 1/1
a.
c
■
• •** >,
XI
o
£ •n
0 c
4> 00 >s in o
Jo in
a
•«H
«o
■o b
0
1-.
to o
& -o3o
OM^v000ONn«O00ON^\000ON^\O
c
C)
1- cn
■ wu
D. ca •*■«c
647
. L
„ m O CTi rg vO (Ti rg g3 CO in on in rg
o tO
■o
rr C r- in in \n r- rr r^ MD ~m
r~ CO T CO rr rg
rr V •<T to to to tO to
CT>
rg
00
rg
r-- ■n
m in
rg rg rg rg
to to
rg
o
rO T rT in to r^ CO lH rr CT.
o t0 CO ct.
U1 M7J ■w■w
CO v0 in
rg
O
in CO
r- CT. tO r-.
to
3 fT. IV)
«T rr r*
CO g=>
to
rg m CO r-
to to rg rg rg
\0
rg
in to to
rg rg rg rg rg
d
l?
^ r-- cr. __•T CO rr to rg ^
O CT. 00 CO r^ r- r-- r- so \D
o U") rr cr. r- CO rg CO v£J U-) <J O CO in fT. to
i/) o rg Ci sO
rr rr to
^
rg Ch CO r- <£> m rr rr tO
rg rg rg rg rg rg rg
rg
rg
rg
rg rg
o
In
0
rr CO to rg to CT> ~
on U1 rg O u-
o rg CTi rr in o r- to r- rg r* to
•
r- o> r-. rr rg Oi CO r^ vO m rr to rg rg M
*y rr to to rg rg rg rg rg rg rg rg rg rg
d >1-. <S
a\ r- rr m to LD g= CO V o% rr LO O o
& rg rr O o rg rr to rg to in on
_. r- rg rg to
•— r-
o m rr u-i
f> C r-~ in to o^ CO r-. g-j un rr to to rg
TT r- to rO to rg IN rg rg rg rg rg CM ^-t
o •a
vC
ct> o 00 m rn O CT, r- tO tO CO C ■O
rr ^i to
o
r-j
00 tO rr en CT, o m rg o
■V rg rr, CO r»
C7i r>j rr on rg
i/l in rr to rg rg
<;
«- -r rr rg rg rg rj rg rg rg eg rg rg rg rg rg
m O
i/>
00 o n to K) L0 O o
vO Ol r- o r» CO
rg LO
r- *y rg
M sC
rg
r-
rr
r- C7> CO CT.
CO rg gO
*—•
g^ g^ rg r- rr
CT. u-i 0O o co r- O lO rr to rg rg O
rr T t tO to to rg rg rg rg rg rg rg rg rg re 5 tn
o " mi
ea
—« tO CTi o tO tO T Oi g^ u->
•T in
M in rg r? rg
o CTi*"g on o LH T «c r>- r*- o\ to r^
%o tO Oi r^ rr rg c fT. r- g^: LD rf tO rg rg
rr *T Kl rO W to rg rg rg rg rg rg rg rg rg
o
l/l
O LTl rO go lO rr
r-. l/l £ O CO cr.
O •^T —rg o
P\ O lo LP, r^
^ O r- CO in in g?
rr.
rr to rr —«
in CT, rr en in r- to
o tO c r-- m rg e Oi
rr rr
o in rr to
rg rg rg rg rg rg
rg rj
rg rg rg rg
Q ene m CT.
00 r*-
o
■DOC
— LTi CO m rT rr tO LO CO rg tn r- r- O O O to to tO I-O tO tO
Kl >o M rg U1 rg rg c: O rg u-i on rg gD g0 rg
o rr r-l C\ o rT rj Q CO r- O in rr to rg o r- r-
v rr to <"M rg rg rg rg rg rg rg
d
re re
pg co an rO tO m Q E .H
O m Oi cr. CO LH TT rr m r- O to r- r-g r~ rg on
o
O
rg
*r
o
rr m
lT rg rT. r* V.*) tO rj rg
rg rg rg r-i rj rg rg r-j rg
■=- en r--
o
sO -* m ct.
rr o g?
U*> "T r- to rr r- —
CC g0 en g^ r- O to r~ —i go —. r- to
o CD r» m CO sD U~> T fO rg rg O r- r-- r-.
o
n rj rg rg rg rg rg rg rg —— >.'*
-/ OO O rg rr go CC C rj rr vO CC o rg rr g= 00 C rg rr o co o rg rr gp
"/ m rg rg rg Kl rO to to to rr rr rr rr rr m in m m u~>
/« -« rg.
/"
648
1.4226 1.3987 1.3783 1.
3606 1.3450 1.3313 1.3190 1.3076 1.2982 1.2892
1.80 of
for
and
>B,
given
Based
"Table
Ppoints
Krivoy
•eby
pRaul
Mr.
cPaonrucrsevonpe,tasrg,ed
sexpressed
by
JCompiled
with
Amos,
O.E.
and
Nixon
Eric
in
L,
N.
m82.
toeanahnsdusaroend,"
1.4076 1.3855 1.3665 1.
3500 1.3354 1.3225 1.3110 1.
3006 1.2913 1.2826
1.70
1.4441 1.4162 1.3927 1.3724 1.3548 3395 1. 1.3138 1.3030 1.2958 1.2845 1.2761
1. 3259
1.60
10*
of
points
sthe
Upper
p)/o;
dt(x
0.90)'
(ct=
aendvairda-izted 1.4258 1.4004 1.3787 1. 38 1.3294 1.3167 3054
3600 1.34 1. 1.2954 1.2910 1.2779 1.2697
1.50
POF
TOINTS
CURVES
ERCAERNTSAOGNE
1.4695 1.4364 1.4089 3856 3656 1.3483 1.3331 1.3198 1.
B459-498,
iNos.
by
Pearson,
S.
50,
aVol.
4,
1963.
E.
3
ntorpp.
mndedutcriokna,
1. 1. 3080 1.2972 1.2880 1.2844 1.2714 1.2635
1.40
CTABLE
15.3
ontin-ued.
1.4598 1.4273 1.4003 1.3775 1.3578 1.3409 1.3260. 1.3129 1.3013 1.2911 1.2812 1.2733 1.2652 1.2583 1.2514
1.20 49i>4
1.
1.5003 1.4588 1.4/48 1.3967 1.3730 1.3528 1.3354 1.3203 1.3066 1.2948 1.2844 1.2749 1.2659 1.2585 1.2515 1.2451 1.2391 1.2337
0.90
2960 1.2850 1.2751 1.2661 1.2582 1.2513 1.2443 1.2383 1.2334 1.2274
1.5170 1.4715 1.4346 1.4042 1.3788 1.3572 1.3388 1.3227 1.3088 1.
0.80
3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4 4.6 4.8 5.0 5.2 5.4 5.6 5.8 6.0 6.2 6.4 6.6
649
s 1 1 1 l 1 1 l i :*» **"* = £ | | . , ; , ,J8 SSSXX S
s 1 | i i 1 1 1 1 8SS sssss s 111 1 i i?5* 3S333 =
^ e •i
s i mil ;Ssr; sssss <
I
5
e
• i | 1 i 1 i :;««? :???? ? £
*i 1 ill'1 I 33SS 3333S &
%
m
}'A«* *> « » r- *- r- r- r- T- X « s
■ ■a>2 1 |*«S£ sr*p* 55332 £ ©
1 '••«««• *••>*>•*« *»*•■* •*
CO *;;o •m»>m XXSSC C
i 3S-- -22*2 - - - Z- 2
[ *• i •:?;?? ssts: sxsss i
o
1 6
c
o
I 8SS? £=:r: = 2£s£ 8
■ i asrss sssss sssss :
i 1 S'S'sSS SSSSS SSSSS I
3
1
, SXSSS SSSSS SSEES 5
3 0
; 11 l«» e
CQ 75 V
/ e
6
■ 1!!*! sxsss !?!!! : 5 |f atl r- »- ^ »- f^ »- *• •• •• • «
£5 { 0
■ If!** sssss SSSSS s 0
, |ttes sssss s2?xa s
■i at
o
1 i|MI XXXX3 SSSSS s e
|SCSS SSSSS SSSSX X
i 6
, •£«•• ■>«••• e , ssrrr sssxs sssss s
0
i 1 sssss sssss SSSSS £
• 0 •
a t
0
1 II3?| |3S3l XMSS I
s
£ St
|
a
; sssss :;;:; « 1 c * S i f a •«««« • X33• oscSc 2 |
650
I <»
\ 1 1 1 1 1 in?? 5*555 \
i in 1 1 i i|bj ;??;? ? j
8 j 1 Mill iii ss3S51S 8
***** *
I 1 1 Mill Mill mil *8
i I
1!
1
8 i inn
l 1 l l 1 1 ijflj 51511 ;
|S*s* essts s |
5! '
. I i
1 1 1 1 1
II II 1
sssss *a
111! *****
iISSS ***** *a
88188
1 s* i in 1 1 I-IU !!!.. 2 15sa? SZ.SS
!|- Mill ***** a*
*1 I MM? 5*fH 88:8s 5
!i ' mil 88:=: 8*888 8
***** ***** *
j s|i iiif* ittts ztaas s
88838 a
i i ■ 1 1 1" ***** •*****»
83:38
4 * 1 -- ---_* ***** *
*A
S ?* I I :**=
•>— — ???=?
••«■•■ «ssjs
MM Ml MM sM
d 6 ***** xasea
2 ! i "2.2 SSSSX ***** a*
CQ I < * 1 |SSCS S=8EZ 3JSSS t 1
8
• ■ J ! - ,1881
1 **** ***** ***** *
i |*?ss
r * ■>*»** =ssss
*M**« sssss
***** *: I ! i ill!: 11111 *>*,*£* *
* i itat: :ss:s sssss s
4 '..'•1 ***** ***** «
: i ' ifffi S888S 88=:: •
t
j
! i sssrs sssss zzsss =
--••I* ***** ***** a
lj ' 2:2:: ***** 1111! 2
! i rszes sssss trees s
* -.*** ***** timil *
Si ' 88.81 11888 SS8ZS I
z
1
i 1 !?.:. :..:. ._::: £ 5 i 1 82888 ■erst 88888 «
* 1 ' ***** MM****
S B S283S ssrss SSSSS 8
e — -mm mm *•«*** *MM** Ml s* :. ;X ***** zsSSSSS 88SS8 SSSSZ
e
| * •SS— ?»!"» S n 2 t, » * c ! ■ tSSSX X«S3S sssss s
1
= z
r- — n •*• a it • n • • — «•>••■ k — 1 • O O * » O ■• r- • - n » • f- ■ *
e 1 ••*(•- ***mm *■*•>« m
S w *••■•» •. n p> c M n r- s © — *. M
651
652 DISTRIBUTIONAL TEST AND REPAIR TIMES
TABLE 15.8
PERCENTAGE POINTS OF PEARSON CURVES
Lower 5% point* of the itandardiaed deviate u. — «■)/», (a - 0.05). Note that for positive akewneas,
lc, mi > 0, the deviate* in thil table are negative.
\8,
0.00 0.01 0.03 0.05 0.10 (1.15 0.20 0.30 0.40 0.50 0.60 0.70
1
1.80
2 2 0 8107 0 7484 |
2.4 0.8944 0 8267 0 7659 0 7109
2 6 0.9792 0 9068 0 8412 0 7820 0 72S2 0 6788 1 1
2 8 1.0608 0 9872 0 9179 0 8545 0 7966 | 0 7440 0 6956 0 6509
3.0 1.1337 1.0632 0.9938 0 9277 0 8663 10 8100 0 7584 0 7110 0 6672 1
3.2 1 1947 1 1304 1.0646 0 9993 0 9362 | 0 8769 0 8222 0 7716 0 7251 0 6820 0 6419
3 4 1.2437 1 1868 1 1269 1.0653 1 0036 | 0 9435 0 8863 0 8332 0 7836 0 7380 ; 0 6957
3.6 1 2824 1 2326 1.1793 1 1232 1 0652 1 1 0069 0 9496 0 8946 0 8427 0 7946 0 7498
3.8 1.3127 1 2692 1.2222 1 1720 1 1192 | 1 0646 1 0093 0 9546 0.9017 0 8514 0 8046
4 0 1.3363 1 2982 1.2569 1.2124 1 1650 1 1.1151 1 0634 1 0109 0 9587 0 9078 1 0 8591
4 2 1.3550 1.3212 1.2848 1.2454 1 2031 ! 1 1581 1 1108 1 0617 1 0117 0 9618 10 9129
4 4 1.3698 1 3400 1.3072 1 2722 1 2346! 1 1943 1.1514 1.1064 1 0596 1 0120 0 9642
4.6 1.3545 1.3254 1.2938 1.2605 1 1.2244 1.1857 1 1448 1 1018 1 0572 1.0117
4.8 1 3403 1.3121 1.2819 1.2494 1.2145 1.1775 1.1383 1 0971 1 0545
5.0 1.3268 1 2995 1.2701 1.2387 1.2052 1.1695 1 1318 1 0924
5.2 1.3142 1 2876 1 2391 1.2286 1 1962 1.1618 1.1255
6.4 1.3022 1.2762 1 2484 1 2189 1 1875 1.1542
6.6 1 2907 1 2653 1.2383 1 2092 1.1790
6.8 1 3030 1 2797 1.2549 1 2285 1 2005
6.0 1.2920 1 2691 | 1 2448 1 2190
6.2 1.2814! 1.2589 1 2351
6.4 1.2711 1 1 2490
6.6 ! 1.2612
HOW TO APPLY 653
TABLE 15.9
PERCENTAGE POINTS OF PEARSON CVRVES
Upper 5% pointa of the lUndardiMd aerial* It. — »)/«. (• - 0.9S).
0.00 0.01 003 0.05 0.10 0.15 0.20 OJO 0.40 0.50 0.60 0.70
*
12 1.1547 1 2056 1.2326 1.2458 1.2579
1 4 1 3191 1.3751 1 4106 1.4271 1.4438 1 4436 1 4348 1.4042
1.6 1 4038 1 3249 1.5618 1 5832 1.6128 1 6249 I 6258 1 6031 1.5604
18 1 5588 1 6151 1.6517 1.6751 1.7138 1.7390 1 7558 1 7687 1.7546 1.7153 1 659S 1 6005
2 0 1.6108 1 6602 1.6941 1 7168 1.7376 1 7881 1.8129 1 8503 1 8721 1.8748 1.8538 1 8078
2.2 1 6361 1 6793 1.7100 1 7310 1.7702 1 8011 1 8279 1.8741 1 9127 1 9426 1 9606 1 9609
2.4 1 6467 1 6849 1 7126 1 7318 1.7682 1 7977 1.8238 1 8709 1.9138 1 9535 1 9888 2 0174
2 6 1 6495 1 6837 1.7088 1.7203 1.7600 1 7874 1 8119 1 8569 1.8991 1 9400 1 9799 2.0181
2 8 1 6483 1.6792 1 7021 1.7183 1 7493 1 7746 1.7975 1.8394 1.8792 1.9183 1.9574 1 996S
3 0 1 6449 1.6733 1 6944 1.7093 1.7380 1 .7616 1 7827 1 8216 1.8585 1 8949 1.9317 1 9690
3.2 1.7488 1.7686 1.8046 1.8388 1.8724 1 9064 1.9410
3 4 1.7890 1.8207 1 8517 1 8830 1 9148
3 6 1 8041 1.8330 1 8618 I 8911
3 8 1.8160 1 8423 1 8698
4.0 1 825S 1 8508
4.2 1 1.8338
!
0.80 0.90 1.00 1.10 1.20 1.30 1.40 1.50 1.60 1.70 1.80
2 0 1 7453 1 6803
2 2 1 9377 1 8886 1 8221 1.7532
2 4 2 0350 2 03S4 2 0119 1.9614 1 8924 1 8210
2 6 2.0531 2.0832 2 1007 2.1022 2 0792 2 0285 1.9580 1.8847
2 8 2 0360 2 0743 2 1100 2 1402 2.1602 2 1632 2 1415 2.0910 2 0197 1 9451
3.0 2 0073 2.0464 2 0859 2 1249 2.1618 2.1935 2 2152 2.2200 2.1997 2 1499 2.0782
3.2 1.9767 2.0136 2 0513 2.0912 2.1314 2 1715 2 2096 2 2429 2 2665 2.2733 2 2547
3 4 1 9476 1 9816 2 0170 2.0640 2 0926 2 1325 2 1736 2 2147 2 2543 2 2893 2 3148
3 6 1.9211 I 9523 1.9847 2 0185 2 0S41 2 0913 2.1304 2 1711 2 2130 2 2554 2 2965
3.8 1.8975 1 9259 1.9555 1.9864 2 0187 2.0526 2 0884 2.1261 2 1658 2 2073 2 2503
4.0 1.8763 1 9026 1 9296 19577 1.9871 2 0179 2.0503 2 0848 2.1207 2 1589 2 1993
4.2 1.8575 1.8817 1 9067 1.9324 1 9592 1 9872 2 0166 2 0475 2 0801 2 1146 2 1511
4.4 1.8407 1.8631 1 8861 1.9100 1 9345 1.9601 1.9868 2.0148 2 0443 2 0753 2 1082
4.6 1.8466 1 8679 1.8899 1 9126 1.9361 1.9606 1.9861 2 0128 2 0409 2 0704
4.8 1.8517 1.8722 1.8932 I 9148 1.9374 1.9608 1 9852 2.0107 2 0374
5 0 1 8562 1 8758 1 5959 1 9166| 1.9383 1 9607 1 9840 2 0084
5.2 1 8602 1.8790 1 8983! 1 918-4 1 9389 1.9606 1 9828
5 4 1.8637 1 8817 1 9003 1 9197 1 9396 l 9003
5.6 1.8670 1.8844 1 9023 1.9203 1 9399
5 8 1.8531 I 8699 1 8867 1.9040 1 9217
6.0 ' 1.8567 1 8725 1.8887 1 9054
6.2 1 1.8596 1 8749 1 8906
6.4 1.8623 1 8771
6.6 i 1 8647
| 1
654 DISTRIBUTIONAL TEST AND REPAIR TIMES
TABLE 15.10
PERCENTAGE POINTS OF PEARSON Ct'RVES
Lower 2.5% point* of the standaniued deviate it. — »)/», (a - 0.035). Note: If mi > 0, the variatea
in thil table are negative.
0.00 0.01 0.03 0.05 0.10 0.15 0.20 0.30 0.40 1 0.50 0.60 0.70
0.80 0.90 ion 1.10 1.20 1.30 1.40 1.50 1.60 1.70 1.80
1
2 2 0 8107 0 74%4
2 4 0 8944 0 5267 0 7659 0 7109
2 6 0 9811 0 9072 0 8412 0 7820 0 72S2 0 8788
2 8 1 0700 0 9903 0 9189 0 8545 0 7966 0 7440 0 6956
3 0 1.1573 1 0747 0 99S5 0 9292 0 8666 0 8100 0 75S4 0 7110 0 6672
3 2 1 2386 1 1568 1 0786 1 0055 0 9385 0 8776 0 8222 0 7716 0 7251 0 6820 0 6419
3 4 1 3112 1 2329 1 1558 1 0815 1 0115 0 9467 0 8874 0 8332 0 7836 0 7380 0 6957
3 6 1 3738 1 3009 1 2272 1 1544 1 0837 1 0166 0 9540 0 8962 0 8433 0 7946 0 7498
3 8 1 4272 1 3600 1 2911 1 2215 1 1524 1 0851 1 0207 0 9602 0 9O40 0 8524 0 8046
4 0 1 4722 1 4106 1 3470 1 2818 1 2158 1 1501 1 0858 1 0240 0 9655 0 9109 0 8603
4 2 1 5100 1 4537 1 3962 1 3348 1 2728 1 2100 1 1474 1 0858 1 0264 0 9699 0 9169
4 4 1 5420 1 4900 1 4366 1 3806 1 3231 1 2640 1 2041 1 1443 1 0853 1 0282 0 9736
4 6 1 4721 1 4200 1 3671 1 3118 1 2555 1 1982 1 1409 1 0844 1 0293
4.8 1 5026 1.4649 1 4055 1 3541 1 3012 1 2470 1.1922 1.1372 1 0829
so 1.4847 1 4389 1.3905 1.3417 1 2908 1.2387 1.1861 1.1333
5 2 1 4679 1.4235 1 3773 1 3297 1 2808 1 2306 1 1800
5 4 1 4519 1 4089 1 3643 1 3183 1 2710 1 2227
5 6 I 4366 1 3948 1 3517 1 3073 1 2612
5 8 1 4611 1 4220 1 3814 1 3395 1 2959
6 0 1 4460 1 4079 1 3685 1.3277
6 2 1 4315 1 3943 1 3559
6 4 1 4176 1 3812
6.6 1 4041
MOH 01 dVAid SS9
31HVI II'SI
30VJA"3DH3d SXNIOd JO \OSav3d S3AH(U>
'»<><'fl %S'Z nned jo »qi pMipjipont w*1"* *x) ~ '•/(* ■) " '(Si6'0
•rS, 00T> io-o ECO SCO 010 sro OZO 0E0 0*0 OSO 090 OiO
TABLE 15.12
PERCENTAGE POINTS OF PEARSON CURVES
Lower 1% pom u of the iUndardixed varute U. — »)/», (a - 0.Q1). Note: If »i > 0, the deviatca
in this Uble *rc n«>u«t
0.00 0.01 0.03 0.05 0.10 0.15 0.20 0.30 0.40 0.50 0.60 0.70
1
1.51) 1.611 1.70 ISO
i
2 2 0 8107 0 7484 i
3iavi crsi
3OVXN30H3J SIXKM JO MOSHV3d S3AHH3
rtddq t % noiod p mt patifuvpma *1«iA»p **) - (" '*/ °) - (660
000 100 £00 900 010 910 020 OEO OVO 090 090 010
z 0 I 99H
z z I 1Z16 I 9Vl>N 1 IZZ8
z » z fr>IZZ I HE Z sl)90 I H>96 I SZ68
I 9 Z 9l£» Z UG9E Z ZECZ z :suz Z 91ZI e:io i 1-996
z a Z 1669 Z IZ99 Z Z
1019 ZEVV Z 1E9E Z91Z Z 1981 E 9001 10ZO
E n Hlt't Z 1E01 Z Z
1919 99E9 Z lusv 101! Z ISZf Mt'I i«
91»Z ZU9I z :8:o
E Z Z 9008 Z IZ08 Z Z
9»61 £911 Z I9H ZZ01 I 6£» z tut b ii»» 69E6 z nvut
£ > Z 0998 Z ZG9S z ita Z WIS Z ii£9k 1Z>8 Z V608 Z 1E91 SZ01 ". :sZ9 Z 9EK
E 9 I 1968 Z GH6 Z 96Z6 Z 88C6 Z IZH E8E6 Z 19Z6 Z 8Z06 1 6:98 S6I8 z ::r:
£ 1 Z 9ZZ6 Z 69*6 Z Z996 Z 1ZS6 z ttlX 9Z00 7.UHJO Z £866 Z 8C86 Z696
* 0 Z COM Z Z196 Z 0166 £ 1ZI0 £ COEO 6HO E 6990 E 9190 £ 9190 EWO £ ►8E0
► Z Z 1E96 Z 8186 E 1100 £ 9ICO £ 9CVO 1E10 S680 E ZEOI £ 11 92 1111 C 6911
Z EZ96 Z ZI66 £ 9810 £ mo E 9890 1160 8111 GUI'S E 99*1 6:si 199
* t £ 99Z0 e can E CSiO EZOI 99ZI t H 89 E Z991 9E81 r.86i
B> e toco C 9190 E i£W) 0601 WEI E 8991 C 6811 ?M1 anz
9 0 C 1090 C 6990 EZII 11EI E EZ9I £ 6981 woz 96ZZ
9 Z £ 9:so I OH OOM 0M'f E V68I SZ1Z 99EZ
9 ^ SEN c i sac BM'I c 9681 StIZ «:n
9 9 £ 18E1 •arc e 9881 ieiz JiK
9 8 •sere E 1191 1 0981 onz 8»CZ
9 0 CKI'C e 0C8I z eioz c ozez
9 Z 8811 c e ccoz e 8:k
U e i86i
9 9 e sit:
658 DISTRIBUTIONAL TEST AND REPAIR TIMES
TABLE 15.14
PERCENTAGE POINTS OF PEARSON CURVES
Lower 0.5% poinu of the Rudsrdiied deviate U. - »)/«, la - O.OOoi. Note: If m > 0. the riritw
is this uble are nettstive
0.00 0.01 0.03 0.05 a 10 0.15 0.20 0.30 0.4U 0.50 0.60 0.70
2 2 0 8107 0 7484
2 4 0 8944 0 82G7 0 7659 0 7109
2 6 0 9816 0 9072 0 8412 0 7820 0 7282
2 S 1 0736 0 9912 0 9189 0.8545 0 6956
0 7966 j 0 7440
3 0 1.1713 1 0799 1 0000 0 9295 0 8666 !0 8100 0 7584 0 7110 0 6672
3.2 I 2735 1 1735 1 0854 1 0079 0 9391 |0 8776 0 8222 0 7716 0 7251 0 6820 0 6419
3 4 1.3776 1 2710 1 1752 1 0901 1 0148 !0 9477 0 8876 0 8332 (1 7836 0 73S0 0 6957
3 6 1 4803 1 3697 1 2682 1 1763 1 0941 | 1 0208 0 9554 0 8966 0 84X1 0 7946 0 7498
3 8 1.5783 1 4667 1 3619 1 2650 1 1768 1 0974 1 0261 0 9623 0 WM6 0 8524 0 8046
4 0 1.6700 1 5596 1.4538 1 3542 1 2616 1 1768 1 0999 1 03OG 0 9682 0 9119 0 8606
4 2 1.7540 1.6465 1 5420 1 4416 1 3465 1 2578 1.1761 1 1017 1 0343 0 9734 0 9183
4.4 1.8305 1 7263 I 6246 1 5254 1.4296 1 3388 1 2536 1 1749 1 1029 1 0373 0.9779
4.6 1 7014 1 6041 1 5094 1 4180 I 3309 1 2491 1 1732 1 1034 1 0397
4 8 1 7718 1 6774 1 5848 1 1 4941 1 4064 1 3230 1 2443 1 1710 1 1034
5.0 1 7453 1 6550 1 5661 1 4793 1 3953 1 3150 1 2391 1.1684
5.2 1.7202 1 6336 1.5483 1 4648 1 3842 1 3070 1 2338
5 4 1 6964 1.6132 1 5312 1 4509 1 3731 1 2987
5 6 1 6737 1 5935 1 5144 1 4373 1 3624
5 8 1 7298 1 6519 1 5746 1 4984 1 4238
6.0 1 7061 1 6309 1 5563 1 4827
6.2 1 6836 1 610S 1 5385
6 4 1 6618 1.5912
11 1 6408
HOW TO APPLY 659
TABLE 15.15
PERCENTAGE POINTS OF PEARSON CURVES
Upper 0.5% point* of Um itudtrdisad demia (z. - »)/•, (a - .995).
0.00 0.01 0.03 0.0S 0.10 0.15 0.20 0.30 0.40 0.50 0.60 0.70
2 0 1 7455
o o 1 9733 1 8056 1 8223
2 4 2 227S 2 1383 2 0515 1.9694 1 8928
2 6 2.4854. 2 3996 2 3069 2.212>J 2 1222 2 0373 1.9484
2 S 2 7151 2 6466 2 5653 2 4741 2 3774 2 2804 2 1875 2 1006
3 0 2 9010 2.8550 2.7950 2 7213 2 6356 2 5409 2 4415 2 3426 2.2483 2 1602
3 2 3 0456 3 0197 2.9815 2 9300 2 8654 2 7879 2 6990 2 6019 2 5008 2 4007 2 3056
3 4 3.1561 3 1460 3 1258 3 0949 3 0422 2 9071 2 9289 2 8484 2.7572 2 6483 2.5562
3 e 3.2406 3 2421 3 2360 3 2213 3 1972 3 1624 3 1163 3 0579 2.9869 2.9041 2.8111
3 S 3 3055 3 3158 3 3200 3.3173 3.3072 3.2890 3.2417 3.2239 3.1748 3.1138 3 0406
4 0 3 3565 3 3726 3 3841 3 3902 3 3909 3.3848 3 3717 3.3510 3.3208 3 2805 3 2291
4 2 3 3962 3.4173 3 4339 3 4484 3 4346 3 4579 3 4446 3 4468 3.4318 3 4083 3.3759
4 4 3 4279 3 4513 3 4724 3 4899 3.5038 3 5137 3 4192 3 5202 3.5155 3 5051 3 4875
4 6 3 5031 3.5241 3 6419 3.3464 3.4677 3.5756 3.5793 3.5782 3.5718
4.8 3 3275 3 5510 3.5714 3.3900 3.6043 3.6183 3 6277 3.6335 3 6354
5 0 3.5721 3 5948 3 6136 3 6342 3 6512 3 6647 3 6760 3.6840
5 2 3 6136 3.6364 3.6469 3 6764 3 6937 3.7084 3.7206
5 4 3.6524 3 6755 3 6969 3 7155 3.7328 3.7493
SI 3 6900 3. 7116 3.7333 3 7528 3.7709
5 8 3.7008 3 7245 3.7467 3.7681 3.7873
6 0 3.7341 3.7572 3.7796 3 8010
0 2 3 7645 3.7885 3.8111
0 4 3.7949 3.8183
6 6 3.8247
660 DISTRIBUTIONAL TEST AND REPAIR TIMES
and
z'n+2 = 1.3406.
Using these values, find
6* = 1.3223 - 2 (1.3041) + 1.2848 = -0.0011,
Sl+1 = 1.3406 - 2 (1.3223) + 1.3041 = 0.0001.
These values are used to calculate
9 = [# - ft (n)] 10,
0 = [(0.27) - (0.20)] 10 ■ (0.07) 10,
or
9 = 0.7.
Substitution of these values into Eq. (15.17) yields
z' = (1 - 6)z'n + 9 ■ z'n+l -±9(1-9) (S2 •< + fi»+1 • ~<+1) ,
or
z* = 1.316908.
The value of z', z" = 1.3169, corresponds to /3J = 3.49, /?j = 0.27,
and a = 0.90.
EXAMPLE 15-2
Given j3j = 4.92, 0J = 0.76, and a = 0.10 find the lower 10%
percentage point, a — 0.10, z value.
SOLUTION TO EXAMPLE 15-2 i
Do the following:
Prepare the following table using linear interpolation:
and
z'n+2 = 1.1384.
Using these values, find
Si = 1.1455 - 2 (1.1519) + 1.1573 -0.001,
Sl+1 = 1.1384 - 2 (1.1455) + 1.1519 -0.0007,
6 = [(0.76) -(0.7)] 10,
and
9 = 0.6.
662 DISTRIBUTIONAL TEST AND REPAIR TIMES
t* = 1.148177.
The value of z', z* = 1.1482, has to have a negative sign because
a = 0.10; consequently, z* = —1.1482, corresponds to /JJ = 4.92,
PI = 0.76, and a = 0.10.
Tables 15.2 and 15.3 were generated using the previously presented
interpolation procedure and the tables given in [4].
The following table provides a comparison of the to.90, *o.95 and £0.99
values obtained using the Central Limit Theorem and the Moments
Method:
Comparison of the results
Central Limit Method of
Theorem, Moments,
hr hr
*0.90 59.1 59.3
<0.95 61.2 62.0
*0.99 65.2 67.4
There is a close correspondence for the 90th and 95th percentiles, but
a significant divergence for the 99th percentile.
The normal pdf approximation is frequently least adequate at the
extreme tails of a distribution. The moments method approximation
"turns out to be better than expected" and the retention of only the
lowest order terms is frequently adequate! The method is limited to
functions to which a Taylor's series expansion about the mean exists
and the partial derivatives are not too difficult to obtain and are mod
erate in number. Numerical methods may be used to find the partial
derivatives.
or
i = f(hih,hiU,h,t6)-
where
i: = 1, 2, • • • , 6 = random variables involved,
and
j = 1,2,- ••, 10,000,- •• = number of trials, or sets of random
numbers generated,
or do as shown in Fig. 15.4.
dcthe
i15.2b
DFig.
usemturilebarutm-ivone
vof
fati
urnicatbiloen
F(l2)
cthe
dufor
PFig.
emr15.2a
utolecaretm-diuvnreing
vof
fdaiursntircaitbuiltoien
t\.
cDiuesmtuerliabmtuinvoen
dthe
vfaute
of rnicatbiloen
Fig.
2d
15.
Fig.
dcthe
Di15.2c
uesmtureilrba-umtinvoe
vfof
a£3.
urnicatbiloen
«,
F(l,)
OS
666 DISTRIBUTIONAL TEST AND REPAIR TIMES
vvalue
ofat\.
of
rt6.
iable
F(t6)
vN-U-«,
Rn-u-
8
Finding
Fig.
s10,000
randomly
15.5a
15.5b
eFinding
lormoreec-t-edd
vt\.
of
values
at6.
riable
t,i,10,000
N-ll-S,
10,000
-•
F(g
',.
HOW TO APPLY 669
Yfl = f(h,N\h,\]---;tn,N),
or
Station number 1 2 3 4 5 6
Random value,
ti, hr 11.3 17.5 1.9 6.3 0.3 9.2
or
fi = 46.5 hr.
0.90(100+ 1) = 90.9th,
ranked observation in 100 simulations. If the 90th observation
was 58.8 hr and the 91st was 59.4 hr, the estimate of the 90th
percentile would be
Similarly, the 95th percentile would be 61.0 hr, and the 99th
percentile would be 70.3 hr. These compare closely with those of
the previous two methods, as shown in the following table:
Percentile - hr
90th 95th 99th
Method 1 59.1 61.2 65.2
Method 2 59.3 62.0 67.4
Method 3 59.3 61.0 70.3
10. Calculate the four moments from the Monte Carlo values of the
output, or of t, and use these in a Pearson distribution approxi
mation; namely, /?i and /?2, or obtain an empirical fit to the his
togram constructed from the Monte Carlo values and find the ap
proximations to the desired percentiles. The moments calculated
from the Monte Carlo values are very close to those obtained by
the moments method, consequently, similar approximations are
obtained.
1. Probability plotting.
2. Maximum likelihood estimators.
3. Matching moments.
4. Least squares.
5. Parameters calculation.
__ _ i _ _
672 DISTRIBUTIONAL TEST AND REPAIR TIMES
35
30
25
E
1 20
10
X
From 35.0 40.0 45.0 50.0 55.0 60.0 65.0 70.0
To 39.9 44.9 49.9 54.9 59.9 64.9 69.9 74.9
Total hours
Monte Carlo estimates have associated error bounds. The larger the
number of trials the more precise the final answer is. Given the error
with which the answer is to be determined the required number of
trials can be determined.
If we want to know the confidence interval about the maintainabil
ity for a test-and-repair time of 60 hr, we can determine it by using the
number of trials involved and binomial distribution confidence bounds.
For example if 94 trials out of 100 gave a total test-and-repair time of
60 hr or less, then,
* = ^ = 0-94- j
If we desire to know the 95% confidence interval about p, then we use
the 95% confidence interval tables for the binomial pdf. Enter Fig.
15.8 with p = 0.94, go to the sample size 100 curves and read off the
95% confidence limits. In this case
!
Ggiven
dsof
various
from
values
random
eitnearinbadutaironds xj
value
random
obtain
Prto
ocedure
y-■
(*H-y\
-l|oge(l
y'
-/?„)
+7.
= **-*»
f.' e+(t+A)«V
'../_ cV
'2
normal,
svalues.
random
and
uniform,
R^,
Ru,
tandard ,{..«.,}
<
\
•el
J
>
/v00,
—00t
y
lJ
''")]),
e{
(L—
/(«)
0<y<oo.
= KiSsfe)]1},
.e{-ib+i
Ae<y<e
+.
Probability
function
density
r~%
1'
1l—
fl,.\
0
<*
</5rt/—*-j
I$je~,»/'1,
/(»)
0<y<oo.
=
ef(y)
<*-W,
A7<i/<oo.
"A
=
[(y_e)»+Asji
y*w
i
-2
/13/J
j^
— r{*>
*'J
fln\ 'J
f(iA
Dito
stribution values
(integral
sbe
imulated
Exponential
Cli -square Lognormal Su
Johnson Sy
Johnson
TABLE
15.16 Gamma
of/?)
P\/
value
random
obtain
rto
ocedure
y'
lowest—
the
is
fc
where
k,
that
such
integer E-i °g,(i-^)>i- /
i#u,
0<
pf, i(
1
RVi
>p,
f
E
E"^+
«Jri
-/?„)]*.
[_loge(l
^ +
7
7,
=
a-r+3-i
3y
where«i
=Sr„-,
i
,/
l
=_ 2,- ,n. 6
l,1= is
and
/fo
d1.
with
normal
the
random
from
i/Z/v
is
sta<rrfi nistmoreeoverrirbvutailon Rfji-
=ueivaabution than
(1,0).
idfrom
When
uniform
the
value
random values
All
dtypical
Rw
is
value
required,
esareorasiaonegnated
1=1
A1=1 *!/'*,=
E
y
rfSrfe^-1
(/^)
i0<y<l.
-y=)" 1. (nMO-p)n-y>
0.1,2,
f(y)
■••,«•
y
=
fPruonbcatbiolnity
density
•-1e-<*)')
h/(»)
0<y<cc.
=
^,
/(»)=**
0,ly
=,2,- .
242-243].
[1,
another
itaken
of
ndpp.
eonependently
CTABLE
o15.16
ntin-ued.
Dito
stribution be
simulated 7)
values
ofn
and
(integral Binomial
Weibull Poisson
Beta
OS 51
676 DISTRIBUTIONAL TEST AND REPAIR TIMES
-x/n
IHIMtligitllllKtulHOHeU tlOtn OHQIlllldOOUtHtHOHttllUDUfltt »USH
pL2 = 0.875,
and
PU2 = 0.975.
Therefore,
P[0.875 < M{t = 60 hr) < 0.975] = 0.95,
whereas for N = 100 trials
M(t = 60 hr) = 0.94.
To increase the accuracy; i.e., to narrow the confidence bounds about
M(t), the number of Monte Carlo trials, JV, should be increased. For
example for N = 1,000 the confidence limits for a CL = 0.95 become
PL2 = 0.925,
and
PU2 = 0.955,
if 940 of the trials gave t < 60 hr, where now
940
P = 1,000 = 0.940.
Then,
P[0.925 < M(t = 60 hr) < 0.955] = 0.95,
or a significantly narrower interval within which the actual maintain
ability lies, with 1,000 Monte Carlo trials.
The process may also be used to determine the number of Monte
Carlo trials which provide the maximum allowable error, E, in esti
mating M(t). To accomplish this specify E, CL, and p', an initial
estimate of p, or of M(t). For example, I want to be 95% sure that my
M(t) estimate will not differ by more than ±0.05 from the final esti
mated value. Assume an initial estimate of p' = 0.80. If 1,000 Monte
Carlo trials are conducted and 800, or 80%, have values t < 60 hr, the
resulting 95% confidence interval for M(t) is
P[0.775 < M{t = 60 hr) < 0.825] = 0.95,
from Fig. 15.8. Similarly, for N = 400
P[0.755 < M(t = 60 hr) < 0.840] = 0.95,
678 DISTRIBUTIONAL TEST AND REPAIR TIMES
where
A, I
P'(l-P')
(15.20)
N
and substitution of Eq. (15.20) into Eq. (15.19) yields
_ ,2 f V f >
E' Z|
1 Q
2
N
Solving for N yields
P'(l-P')
N £2 zl'Ia . (15.21)
for
CL = 0.95 = 1 - a; a = 0.05; | = 0.025,
21-0.025 = 20.975 = 1-960,
and for
Q
CL = 0.99 = 1 - a; a = 0.001; - = 0.005,
^1-0.005 = 20.995 = 2.576.
or
N £ 246.
This approximation is usually adequate for N p' or N (1 — p') equal to
or greater than 5. If a one-sided bound is needed, instead of two-sided
bounds, then only the difference of the upper (or lower) curve from
the diagonal line of Fig. 15.8 is used to find N. In the equation for N
use zX-a instead of zx_s. , or 1.282 for 90% CL, 1.645 for 95% CL and
2.326 for 99% CL. It must be noted that in finding N, an estimate of p,
or of M(t ), was required, which is the very quantity to be determined
by the Monte Carlo study. This is because the size of the confidence
interval is a function of p, or M (t), as can be seen from examining
Fig. 15.8 and Eq. (15.21) for N. The largest sample size, however, is
required when p = 0.5. Therefore, if p is not known ahead of time, use
p' = 0.5 to get the most conservative sample size. Or conduct some
Monte Carlo trials, get an estimate of p', and use this to obtain N and
the additional trials to get the desired accuracy or error bounds.
To find N for a desired error, E, in estimating the mean of the
output, or t, with a desired CL = 1 — a, the estimate of the standard
deviation, at is needed. Then,
2i-f °t
N = (15.22)
E = ±z1_°2 4L.
y/N
V \ V )
where
P = 2.0, tj - 3.0 hr and 7 = 0.5 hr.
2. Station 2
t at'V2 tt
where
V = 0.5 loge hr and ert< = 1.0 loge hr.
3. Station 3
where
/? = 6 and 77 = 4 hr.
ERROR BOUNDS 681
Do the following:
1. Find the time to repair-and-checkout for the maintainabilities of
90%, 95% and 99%, using the Monte Carlo Simulation method,
with 100, 1,000 and 5,000 t values.
2. Discuss the simulation results.
_L_
682 DISTRIBUTIONAL TEST AND REPAIR TIMES
20 -
-^- Range » 4. 4 73
15 Range ■ 1.352
Range • 2.107
E
J
Standard
deviation • 0.626
-— Mean • 10.322
I Mean ■ 1I.311-' j Mean 10. 903 <—L-r-
10
1 * Standard
—
—
*— deviation
» 0.473 - A
i
n—
5-
90th percentiles
Fig. 15.9 - The plot of the simulation results for the 90th
percentile, in terms of the mean, range and stan
dard deviation for Example 15-3.
ERROR BOUNDS 685
Range • 2.086
/■Standard
viation * 0.657
1
* *-Slean ■ 13.554
Standard
deviation»0."0
9Sth percentiles
Fig. 15.10 — The plot of the simulation results for the 95th
percentile, in terms of the mean, range and stan
dard deviation for Example 15-3.
686 DISTRIBUTIONAL TEST AND REPAIR TIMES
N ■ 100 N - 5,000
Standard
deviation ■
3:
X
X
Meaji.:S.J51
\
z
:s
v~?e ' 5 . C-3
Range ■ :.94J
X
X
1
L
rL- Standard
X deviation » C.339
X
^- Mean - x:i.0S6
Standard
1
i
4c Mean » II. *T
20 deviation » ^ I ■ SCO
*—
99th percentiles.
Fig. 15.11 - The plot of the simulation results for the 99th
percentile, in terms of the mean, range and stan
dard deviation for Example 15-3.
PROBLEMS 687
cmM,a
~ Vp(i-p).
/(<*,)
i.e., emax,a depends on the value of p and of f(a) at ap.
4. The larger deviation of the 99th sample percentiles is due to
the larger value of \/0.99 (1 - 0.99)//(o0.99), because /(ao.99) is
smaller.
PROBLEMS
«=i
where t{ is the respective repair-and-checkout time in each sta
tion. The repair-and-checkout-time distributions for each one of
the four stations are the following:
688 DISTRIBUTIONAL TEST AND REPAIR TIMES
1. Station 1
9E{U) = fuT" '»
where
H = 2 repairs/hr.
2. Station 2
V \ V )
where
fi = 2.5, t? = 3.0 hr, and 7 = 0.5 hr.
3. Station 3
9LN(h) = I* (¥)'
^3 #{' V 2 7T
'3
where
<;3 = 0.6 loge hr, and at< =1.0 loge hr.
4. Station 4
where
/3 = 8, and r? = 5 hr-1.
Do the following:
1. Find the time to repair-and-checkout for the maintainabili
ties of 90%, 95% and 99%, using the Central Limit Theorem
method.
2. Same as in Case 1, but using the Monte Carlo Simulation
method, with 100 and also with 5,000 simulated T values.
3. Discuss comparatively the results of Cases 1 and 2.
4. What is the required number of Monte Carlo Simulations to
determine the 90% maintainability with error of ±0.01 and
90% confidence?
5. What is the required number of Monte Carlo trials to deter
mine the mean time to repair-and-checkout the trucks with
an error on the mean time of ±10% and 90% confidence?
In all cases above write out all equations used and give the step-
by-step procedures used to get your results in addition to the
computer programs used.
PROBLEMS 689
where
P = 2.5, J] = 3.0 hr, and 7 = 0.5 hr.
690 DISTRIBUTIONAL TEST AND REPAIR TIMES
3. Station 3
«,
9LN(<3) =
» -»(¥"
^=e V 3
'3 °V V ^ *
where
i'3 = 0.1 loge hr, and at> = 0.2 loge hr.
4. Station 4
where
/? = 8, and t? = 5 hr-1.
Do the following:
1. Find the time to repair-and-checkout for the maintainabili
ties of 90%, 95% and 99%, using the Central Limit Theorem
method.
2. Same as in Case 1, but using the Monte Carlo Simulation
method, with 100 and also with 5,000 simulated T values.
3. Discuss comparatively the results of Cases 1 and 2.
4. What is the required number of Monte Carlo Simulations to
determine the 90% maintainability with error of ±0.01 and
90% confidence?
5. What is the required number of Monte Carlo trials to deter
mine the mean time to repair-and-checkout the trucks with
an error on the mean time of ±2% and 90% confidence?
In all cases above write out all equations used and give the step-
by-step procedures used to get your results in addition to the
computer programs used.
15-4. Using the simulated total times to repair-and-checkout trucks of
Problem 15-3, do the following:
1. Calculate the four moments two ways:
1.1 Fit the normal, lognormal, Weibull and the extreme-
value-of-the-maxima distributions to the first 100 simu
lated total times to repair-and-checkout the trucks, i.e.,
find thir parameters. Any good method is acceptable!
Then calculate all four moments using these parame
ters.
REFERENCES 691
REFERENCES
1. Hahn, G. T. and Shapiro S. Samuel, Statistical Models in Engi
neering, John Wiley k Sons, New York, N Y, 355 pp., 1967.
2. Lindgren, B. W., Statistical Theory, The MacMillan Company,
Collier-MacMillan, London, 521 pp., 1968.
3. Cramer, H. Mathematical Methods of Statistics, Princeton Uni
versity Press, Princeton, 575 pp., 1951.
4. Krivoy Raul, based on Table of Percentage points of Pearson
Curves for Given y[$ and f32, Expressed in Standard Measure,
Compiled by Johnson, N. L., Eric Nixon and Amos, D. E. with
an introduction by Pearson, E. S., Biometrika, Vol. 1, Table 41,
1954.
5. Pearson, E. S., and Hartley, H. 0., Biometrika, Tables for Statis-
titians, Cambridge University Press, Cambridge, England, Vol.
1, Table 41, 1954.
6. Kececioglu, D., and Jiang, J., Proceedings - Institute of Environ
mental Science, Error Bound Estimation On Monte Carlo Sim
ulations, pp. 124-128, April 23-27, 1990.
692 APPENDIX 15A
APPENDIX 15A
PROOF OF EQ. (15.23) [2, p. 367]
Consider a sample of size M from a continuous distribution, with
cdf F(x) and pdf f(x). Let ap denote the percentile of order p of the
distribution, i.e., the root of the equation F(ap) — p, where 0 < p < 1.
We assume that, in some neighborhood of x = ap, the pdf, f(x) is
continuous and has a continuous derivative f'(x).
(M - I) f(x) dx.
Therefore,
or
y = \/JS/(Q')|{,">-0>1' 15/1.2
then
(M) _y />(!-» 1 + ap. 15A.3
G+1 M /(ap)
The pd/ of Y can be expressed as
«,M»
W+l
»<»)-#nMf^+«,
M /(ot) <fy
15A.4
where
<> P(l-P) 1
<fy M f(apy
and <£(•) is defined by Eq. (15A.1). Substitution of Eq. (15A.1)
into Eq. (15A.4) yields
^-Tcbnp* {IM<
"J^^
694 APPENDIX 15A
M-(l+i)
I p (1 - p) y
f i ./p(1~p) y i_„ i
or
AW = ^,/Lllz£)^(w.,)
/(o„) V M ^Z
•[F(z)]'[l - F(z)]w-('+1)/(2),
where
/p(l -p) y
Then,
&(y) =
^ >->>--*£?
M-(/+l)
/(*)
or
/(a.) (WW)
MJ/) = M ■ M ■ A3.
p(l~p)M-l M\ , M_,
Al = \l M l-p[l)p{1-p) ■
/(*)
f ("p + V^P/T^
An = -1,
f(°p) /(«p)
APPENDIX 15A 695
and
F{z) 1 - F(z) M-(/+l)
A, =
P 1-P J
If F(ap) = p and F'(ap) = /(ap), expanding ^(z) near j/ = 0,
yields
/p(i-p) y
F(z) = F aP +
M f(ap)
^)=-/¥-^^^»(i)
Substituting this into A3 yields
A3 =
M-(i+l)
£— ^ipO^lO^^-f.-^
(1 - p) M 2 PM /2(ap) •(i)
or
A3 = /•//,
AsM-oo,PM^!,M-(/+ 1)— M - / -. (1 - p) M = n.
Let
_ 1 _
696 APPENDIX 15A
or
/ = e{'>/3F»+iMl-F)^f»-i(l-F)»
P (1-P)3/2 /'(°p) „3_I P2 (1-P)2 [/'(°p)'
and
Hence,
As = I II,
_ e[i v^ v-n vT H o-p) vM p y2+°(^)]
[v/p M (1-p) v-v/(1-p) M p r-i 3/2+o(^!?)]
or
i
4s- e(-"2 v2).
Therefore
1
Ky)-- = e" *•
REFERENCES
1. Lindgren, B. W., Statistical Theory, The MacMillan Company,
Collier-MacMillan Limited, London, 521 pp., 1968.
2. Cramer, H., Mathematical Methods of Statistics, Princeton Uni
versity Press, Princeton, 575 pp., 1951.
Chapter 16
697
698 THE WEIBULL PROCESS OF REPAIRABLE UNITS
Condition 2 is
N(t, t+At), completely independent of the failures that occurred in
the interval [0, t], meaning that the repair action restores the equip
ment to its original age condition as just before the failure. Conse
quently, as illustrated in Fig. 16.1, the failure rate of the equipment
is restored to the value it had just before failure through the repairs
performed after failure. It is to be understood that the failure is due
to the surfacing of a latent which the repair action is now removing
thus resulting in the continuing decrease of the failure rate.
Condition 1 says that in a small time interval (t, t + At), a failure
occurs at most once, and the probability of the failure is in proportion
to \(t) and it is time dependent. Condition 2 says that the failures
after time t are independent of the failures that occur before time t.
It has been proven that Ti , the times to the first failure, are Weibull
distributed with pdf
by Condition 2; i.e., the distribution of TT, given the times of the first
{r — 1) failures only depends on the time of the (r — 1) failure. Then,
the conditional reliability of surviving by TT, given it got repaired after
the failure at tT-\, or for Tr, and given that Tr_i = tT-\, Eq. (16.6)
becomes
R{Tr\Tr-i=tr-i) = e { « ; K " , (16.8)
THE WEIBULL PROCESS AND ITS CHARACTERISTICS 699
u x failure point
« E
a«
•> *o o end of repair
£ B
>- a
* 0
ou 5o
■M
09
•■* B
<s
Accumulated operating and repair times
or
f(Tr\ Ti = h,T2 = i2(- • -,Tr_! = xr_!)
(16.11)
where
j("«-(l^l)'ffr-7+fr(l + i).7+|r(i).(i«.u)
In the second integral let z = ( r~'Y)^, then
-*f V v J ' fi
(16.13)
where
/(i;n)= Te-'t"-1 dt
Jo
is the incomplete gamma function, and T(n) is the gamma function.
Therefore, from Eqs. (16.11), (16.12) and (16.13) the mean time be
tween the (r - 1) and the rth failure, given the times to the first (r - 1)
failures, becomes
1; dTT- I dTT
-JHFfUW-,-}, V I? J ' /J
or
tr-l-7
V ( -7|\' 1
mr = -e (16.14)
'{'G)-'[(^
The values of the gamma function, T(n), and of the incomplete gamma
function I(x;n) in Eq. (16.14), may be found in mathematics hand
books, and usually double linear interpolation has to be used. If a
numerical integration subroutine is available, then a more accurate
result can be obtained by integrating Eq. (16.10) directly.
or
E(r) = n (t——ty . (16.15)
or
j*,.*-*,.*
log10 Th - log10 1 jj
(16.18)
In this case, using the values ji/n — 1 with Tj1 = 9.7, and jifn = 10
with Th = 900, yields
0= lQ6io 10 - lQ6io I = Q.5083.
log10 900 - log10 9.7
Therefore, the failure rate function is
0.5083 / T N0-5083-1
or
X(T) = 0.1605 71-04917.
2. From Eq. (16.7) the conditional reliability for T15 = 2,400 hr
and t = 10 hr is
R[Tl6 = (2,400+10) | r, = <!,-• -,T15 = 2,400]
= e -16.5007+ 16.4659
or
R[T16 = (2,400+10) | r15 = 2,400] = 0.9658.
CO
. o
* T-
<J> ■
£
.•^ ■—4
u
_c
o
CO
V\ t-
d
3
£
a rth
0 3
a.
r- -,o f*» o
o) £
H •u
p-
>—
o
3E
i
i
T-
P o M
m CD
f F
[ = u 'u/f -iiun »d s3jn]iB4 jo jsquinu 33bj3ab 3i)X .5°
705
706 THE WEIBULL PROCESSES OF REPAIRABLE UNITS
3. From Eq. (16.14) the mean time between the 15th and the 16th
failure is
m16 = —
9 7 A/ 2,400
e^
\0.5083 \
9.7 )>
f
{T I
\
- /
M()o\0'5083 1
16 0.5083 1 V0.5083.' 9.7 ) ' 0.5083
where
r f- --I = r(1.9673)
V0.5083/ v ; = 0.9866126,
and
r/2,400\0-5083 1
= 7(16.4659; 1.9673) = 0.9866114.
V 9.7 ) ' 0.5083
Therefore,
m16 = (2.70208 X 108)(0. 9866126 - 0.9866114) = 326.50 hr.
4. The rehability for a mission of 10 hr starting at age zero, from
Eq. (16.6), is
^log^-^-log^-T)""1' (16'22)
and 77 from Eq. (16.20) by substituting this /? in it and solving for 77.
EXAMPLE 16-2
Use Graphical Method 2 to estimate the parameters of the Weibull
process represented by the data given in Table 16.1.
SOLUTION TO EXAMPLE 16-2
The average cumulative failure rates listed in Column 3 of Table
16.1, are calculated from
where
j = number of cumulative failures up to the accumulated test
time, Tj.
_
The paired data, (Tj,Xj), are plotted as in Fig. 16.3 yielding an
acceptable straight line, hence 7 = 0. Pick two points on the straight
line, as indicated in Fig. 16.3, and use Eq. (16.22) to determine pa
rameter /? from
_ log10(0.0034)-log10(0.033)
P ~ log10(10,000)-log10(100) + " °-5055-
pH
E
I
eu
o
i
o
s-
E
V
*j
w
I
c
E
ad
g
£ (X
P D
JS
708
THE ESTIMATION OF A WEIBULL PROCESS 709
In this case, for T = 10,000 hr and A(T = 10,000 hr) = 0.0034 fr/hr,
/io^oo0-5065-^05"
' \ 0.0034 J
The results are close to those obtained in Example 16-1 using Graph
ical Method 1.
■n\ n J
The joint density function of the total of r* times to failure, (T\, • • • , Tr»),
which occur during the test period [0, t], is the product of the individ
ual conditional density functions [2], or
r=l
or
(16.24)
(16.25)
710 THE WEIBULL PROCESSES OF REPAIRABLE UNITS
+ £loge(Tr-7) = 0 (16.26)
and
»0*i)__^+/'?hz2y,£=o. (16.27)
dr) t\ \ r) J rj
(■!*LZl\ = r\ (16.28)
or
a- rl (16.29)
But, the estimates of 0 and 7) given by Eqs. (16.29) and (16.30) are
biased. An unbiased estimate of (3 is given by [3]
EXAMPLE 16-3
Find the maximum likelihood estimates of the parameters of the
Weibull process represented by the data given in Table 16.1.
SOLUTION TO EXAMPLE 16-3
From Eqs. (16.29) and (16.31)
16 15 =0.5441,
P- 14
—u
£ioge ( 2aoo\ 27.569
r=l
and
•1) _ 0.5441(15--1)
0tu = 0.5078.
r* 15
From Eq. (16.32)
2,400
f*- = 11.59 hr.
150.5078
Syst 2m 1 System 2
Cumulative Cumulative Cumulative Cumulative
number of test number of test times
failures, times, failures, to failure,
3 Tj, hr 3 Tj, hr
1 1.30
2 3.50
1 5.75
2 8.50
3 10.70
3 14.55
4 16.65
4 28.25
5 31.55
5 35.30
6 36.50
7 38.85
6 46.95
8 47.75
7 49.05
9 50.55
10 66.00
8 71.10
11 73.85
12 74.50
13 83.60
9 95.35
10 96.50
11 99.35
14 125.95
15 141.25
12 143.05
WHEN TESTING MORE THAN ONE UNIT 713
Graphical Method 1
The pooled and ranked data, given in Columns 2 and 3 of Table
16.3, are plotted on Fig. 16.4. It must be pointed out that in Table
16.3 the Column 2 values are j/n, but in this case n = 2. Therefore,
j/2 is used based on Eq. (16.17). In Column 3 the pooled clock hours
of the times to failure of both systems are used.
From Eq. (16.17), the value of the parameter n is the value of Tj
when j — n. Then j/n = 2/2 = 1. In this case, for j/n = 1,
77 = 3.0 hr.
Using Eq. (16.18) and the values j\ = 1 with 7^ = 3.0, and j2 = 10
with Th = 78, yields
log1010 - logipl
0= log1078-log103.0'
or
0 = 0.7067.
Then, the failure rate function is
0.7067 / r \07067-1
A(T) ~ "3J" \JS)
or
-0.2933
\(T) = 0.2356
(n)
where the test time, T, is the accumulated test time to failure for each
system.
714 THE WEIBULL PROCESS OF REPAIRABLE UNITS
I::::
— ^
rtfiH
78
hr 10'
ir1
fflPT^ T,
fTime
hr
ato
ilure,
10
,^
3.0
hr
ti
=
1
Method
Graphical
1.
using
y<£-—
0.1
10 2-—
10 0.1
II ~r •a I
en
716 THE WEIBULL PROCESS OF REPAIRABLE UNITS
Graphical Method 2
From Eq. (16.15) the average number of failures of n systems in
time interval [7, T] is
^"(^T2)''
and the average cumulative failure rate, starting from 7 up to time T,
is
T? = (16.35)
A(Tx)
The average cumulative failure rates are calculated from
\(Tj) = —± -, (16.36)
n(T} - 7)
WHEN TESTING MORE THAN ONE UNIT 717
where
j = total number of failures up to time Tj,
and
n(Tj — 7) = unit-hours of test time accumulated by n sys
tems in time interval [f,Tj].
Here 7 is an unknown parameter. However, 7 may be assumed to be
zero for the initial plot. If the X(Tj) versus Tj points fall acceptably
well on a straight line, then the assumption 7 = 0 is confirmed. If
the X(Tj) versus Tj points do not fall on a straight line, then there are
two possibilities. The first possibility is that the test-analyze-fix-test
process is not a Weibull process, and the second possibility is that 7 is
not equal to zero. These will be discussed later.
In this example, when 7 = 0, the cumulative failure rates, using
Eq. (16.36), are calculated from
- log100.500 - log100.100
13 = log100.6-log10158 +1 = 0-7112>
(n f<0.7112-l\ 0^112
j_
y/
k
10'
Time
T,
failure,
hr
to
10
Method
Graphical
2.
using
K
^
h \i—
0.6
v
\
0.1
103-, 10- 1-
"0.5
fr/hr
"-102-
fr/hr
1
0
£ "o i
00
M
WHEN TESTING MORE THAN ONE UNIT 719
P=—i
i ioge {%)
and
or
12
A= = 0.6412,
18.714
and
143.05
m = 2.968.
120.6412
or
15
fo = 18.7252 = 0.8011,
720 THE WEIBULL PROCESS OF REPAIRABLE UNITS
and
fc„iJi2L„u»72.
15o»on
The unbiased estimates of (3 and r\ for System 1 are,
t}{I = ——j—
143.05 = 2.0870.
_ nMn
120 5878
For System 2
£ = L__Ji(o.8011) = 0.7477,
15
and
4 » 2£^ = 3.7761.
1507477
and
?= ^j = 2.0870 + 3.7761 =
' 2 2
Therefore,
0:6678 /_T_Y0.6678-1
v ; 2.9316 \2.9316^
or
/ T \ -0-3322
A(T)
v ; = 0.2278 V2.9316/
(-—— )
.9316,
3. The predicted value of the failure rate of this system at 500 hr,
using the results in Case 1, for Graphical Method 1, is
/500\~0-2933
\{T = 500 hr) = 0.2356 f — J
THE COMPOSITE WEIBULL PROCESS 721
or
A(r = 500 hr) = 0.05254 fr/hr = 52.54 fr/103 hr,
and for Graphical Method 2 is
/ "ifin \ -0-2888
A(T= 500 ta). 0.2181 (_) ,
or
X(T = 500 hr) = 0.05099 fr/hr = 50.99 fr/103 hr.
4. The predicted value of the failure rate of the system at 500 hr,
using the results in Case 2, or the MLE's, is
-0.3322
X(T
v = 500 hr)' = 0.2278 V2.9316/
f^^^
or
A(T = 500 hr) = 0.04132 fr/hr = 41.32 fr/103 hr.
o
m
o
CD
722
THE COMPOSITE WEIBULL PROCESS 723
and after time T\ the failures follow a Weibull process with the failure
rate function
«-«
The parameters of these two Weibull processes can be estimated di
rectly from the plot in Fig. 16.6. First two straight lines are drawn to
fit the data. Then, the intersection of these two straight lines, corre
sponding to time T\ = 2, 400 hr, is the watershed of the time domains
of the two Weibull processes. Now consider the two processes sepa
rately, and use the method of Section 16.2.1 to obtain the parameters
as follows:
For the first process, pick two points, say, (7\ = 10, ji/n = 0.1)
and (T2 = 1,650, j2/n = 0.7). Then, from Eq. (16.18),
or
_ log100.7-log100.1 _
^-log^MSO-iog.oio"0-3811'
and for the second process, pick two points, say, (T3 = 2, 650, j'3/n =1)
and (T4 = 6, 000, j4/« = 6.8). Then,
log10 6.8 - log10 1
h= = 2.3457.
log106,000-log102,650
The parameters rji and r^ may be found from the plot directly, for
j/n = 1, yielding
??! =4,000hr,
and
% = 2, 650 hr.
Thus, the composite Weibull process is determined. Its failure rate
function then becomes
0.3811-1
0.3811
A(T) =
4,000 U.oooy
or -0.6189
A(T) = 9.5275 X 10- (j£g) for 0<T<2,400,
and
724 THE WEIBULL PROCESS OF REPAIRABLE UNITS
2.3457 / T \2-3457-1
v ' 2,650 V2,650y
or
1.3457
\{T) = 88.517 X 10-5 f—j 1 , for T> 2,400.
PROBLEMS
failure occurred at 1,960 hr, and the mean time to the first
failure.
(4) Same as in Case 3, but using the results in Case 2.
(5) Find the reliability for a mission of 10 hr after the 10th
repair at time 1,960 hr, and find the reliability for a mission
of 10 hr starting from age zero, using the results in Case 1.
(6) Same as in Case 5, but using the results in Case 2.
Syst<em 1 System 2
Cumulative Cumulative Cumulative Cumulative
number of test times number of test times
failures, to failure, failures, to failure,
3 Tj, hr 3 Tj, hr
1 11.5
2 30.0
1 57.5
2 84.6
3 111.7
3 137.5
4 173.5
4 229.0
5 282.0
5 305.2
6 327.5
7 370.6
6 419.5
8 422.4
7 475.0
9 551.5
10 660.0
8 701.5
11 742.1
12 748.5
13 834.0
9 923.4
10 950.2
11 1,005.5
14 1,150.2
15 1,381.7
12 1,392.5
PROBLEMS 729
System 1 System 2
Cumulative Cumulative Cumulative Cumulative
number of test times number of test times
failures, to failure, failures, to failure,
3 7j,hr i Tj, hr
1 140
l 200
2 300
2 380
3 590
3 670
4 720
4 740
5 800
5 880
6 1,020
6 1,070
7 1,130
7 1,270
8 1,280
8 1,340
9 1,410
9 1,540
10 1,670
10 1,800
PROBLEMS 731
Syst em 1 Syst em 2
Cumulative Cumulative Cumulative Cumulative
number of test times number of test times
failures, to failure, failures, to failure,
J Tj,bi 3 Tj, hr
1 180
1 260
2 380
2 490
3 790
3 870
4 950
4 950
5 1,040
5 1,130
6 1,270
6 1,330
7 1,350
7 1,470
8 1,630
8 1,650
9 1,690
9 1,830
10 1,960
11 2,080
10 2,160
12 2,280
732 THE WEIBULL PROCESS OF REPAIRABLE UNITS
System 1 Syst am 2
Cumulative Cumulative Cumulative Cumulative
number of test times number of test times
failures, to failure, failures, to failure,
J Th hr 3 rif hr
1 100
1 140
2 220
2 270
3 420
3 480
4 510
4 530
5 570
5 640
6 730
6 730
7 780
7 810
8 910
8 930
9 980
9 1,010
10 1,130
10 1,190
11 1,210
12 1,320
734 THE WEIBULL PROCESS OF REPAIRABLE UNITS
Syst<2m 1 Syst<-m 2
Cumulative Cumulative Cumulative Cumulative
number of test times number of test times
failures, to failure, failures, to failure,
J Tit hr 3 Tj, hr
1 260
1 310
2 460
2 620
3 680
4 750
5 860
3 1,060
6 1,140
4 1,160
5 1,370
6 1,550
7 1,640
7 1,860
8 1,970
9 2,050
8 2,120
10 2,370
9 2,420
11 2,520
10 2,600
12 2,750
11 3,340
PROBLEMS 735
Syst em 1 Syst<am 2
Cumulative Cumulative Cumulative Cumulative
number of test times number of test times
failures, to failure, failures, to failure,
3 Tit hr J Ij.hr
1 190
1 230
2 340
2 450
3 500
4 560
5 640
3 770
4 850
6 860
5 1,000
6 1,140
7 1,210
7 1,360
8 1,450
9 1,510
8 1,560
10 1,750
9 1,770
11 1,860
10 1,910
12 2,040
11 2,440
12 2,540
736 THE WEIBULL PROCESS OF REPAIRABLE UNITS
and the repaired system is tested again. Assume that the repair
time is negligible and this test-fix-test process follows a Weibull
process based on the test times of each individual system. Do
the following:
(1) Determine the parameters of this Weibull process using the
two graphical methods.
(2) Determine the parameters of this Weibull process using their
maximum likelihood estimates.
(3) Predict the value of the failure rate of this system at a total
test time of 1,600 hr, using the results of Case 1.
(4) Same as in Case 3, but using the results of Case 2.
16-10. Two prototypes of an electronic system are tested simultane
ously and the cumulative times to failure are recorded separately
for each system with the incorporation of repairs. The cumulative
test time at each failure for each system is given in Table 16.13.
Assume that after each failure a repair is performed immediately,
and the repaired system is tested again. Assume that the repair
time is negligible and this test-fix-test process follows a Weibull
process based on the test times of each individual system. Do
the following:
(1) Determine the parameters of this Weibull process using the
two graphical methods.
(2) Determine the parameters of this Weibull process using their
maximum likelihood estimates.
(3) Predict the value of the failure rate of this system at a total
test time of 900 hr, using the results of Case 1.
(4) Same as in Case 3, but using the results of Case 2.
PROBLEMS 737
Syst em 1 Syst am 2
Cumulative Cumulative Cumulative Cumulative
number of test times number of test times
failures, to failure, failures, to failure,
J rjfht 3 Tjthx
1 150
1 170
2 260
2 340
3 390
4 430
5 500
3 590
4 640
5 760
6 660
6 860
7 920
7 1,030
8 1,110
9 1,160
8 1,180
9 1,330
10 1,340
10 1,450
11 1,830
12 1,920
738 THE WEIBULL PROCESS OF REPAIRABLE UNITS
REFERENCES
1. Gisela, Hartler, "Graphical Weibull Analysis of Repairable Systems,"
Quality and Reliability Engineering International, Vol. 1, pp. 23-26,
1985.
2. Crow, L.H., "Confidence Interval Procedures for Reliability Growth
Analysis," Army Material systems Analysis Activity Technical Report
197, AD-A044788, Defense Technical Information Center, Alexandria,
VA, 1977.
3. Tsung-Ming, T. Lin, "A New Method for Estimating Duane Growth
Model Parameters," Proceedings Annual Reliability and Maintainability
Symposium, pp. 389-393, 1985.
Chapter 17
RELIABILITY OF COMPONENTS
WITH A POLICY OF I
REPLACING THOSE THAT FAIL
BY A PRESCRIBED OPERATING
TIME
These are replaced by fresh ones and they operate n cycles thereafter.
The number of these components that will fail after n additional cycles
of operation, using Eq. (17.1), would be
739
740 RELIABILITY OF COMPONENTS
or
NS-NR(ni + n) = N0 Rfa + n). (17.10)
Out of the N0 that started out, the number of those that survive, after
replacement of those that failed by n\, is given by the sum of Eqs.
(17.8) and (17.10), or
Ns-n(n, +n) = N0 {[1 - R{nx)] R(n) + fiL + »)}. (17.11)
Therefore, the total combined reliability of such components, under
the replacement policy considered, is
Ns-RJni + n)
Rt-r(ti-i + n) = (17.12)
Nn
Substitution of Eq. (17.11) into Eq. (17.12), and simplification, yields
RT-M*i +n) = [1 - JJ(m)] R(n) + R(m + n). (17.13)
A study of Eq. (17.13) reveals that the total combined reliability of such
components, under the replacement policy considered, is given by the
probability that either the components fail by n\ cycles, are replaced,
and function successfully for n cycles thereafter, or they do not fail by
ni cycles and thus function successfully the full (n\ + n) cycles.
In Eqs. (17.7), (17.8) and (17.10) the quantities Rfa), R(n) and
R(n\ + n) need to be calculated.
If it has been established that the times-to-failure distribution of
such components is Weibullian, then
Rim) = e ( n ) , (17.14)
i\e
R(n) = e"(V) (17.15)
and
£(ni + n)
R{n1,n) =
J2(»i) '
or
("1+"-1)0
e v i '
R(nun) = (17.17)
e *
742 RELIABILITY OF COMPONENTS
EXAMPLE 17-1
Identical aluminum spindles are operating at an alternating stress
level of 25,000 psi. Their cycles to failure distribution is lognormally
distributed with parameters n' = 5.827 and uni = 0.124. These param
eters are determined using the logarithms to the base 10.
1. If 1,000 such fresh spindles are operating, how many will survive
after 500,000 cycles of operation and how many will fail?
2. If the failed spindles are replaced by new ones and all of them
operate for n = 330,000 cycles thereafter, how many of the fresh
ones will fail?
3. Of those that survived ni cycles, how many will fail during the
additional n cycles?
then,
oo
R(n) = J 4>(z)dz,
z(n')
and
log10330,000-n'
z(n')
<V
or
5.518514-5.827
z(n') = -2.48779,
0.124
and
00
From Case 1,
oc
R{rii + n) = / 4>(z)dz,
*(m+n)'
, , v logio(ni +n)-n'
z(ni + n) = ,
On'
or
= 0.742565.
Therefore,
oo
#(500, 000 cycles + 330, 000 cycles) = / <j>{z) dz = 0.228872.
0.742565
= 849(0.730447),
or
R(n-i + n) — / cf>(z)dz,
*(ni +n)'
_ 5.91908-5.827
0.124
or
2(830,000 cycles)' = 0.742565.
Then,
oo
since
22.9% < 37.9%,
or #(830,000 cycles) with replacement is 65% larger than with
out replacement!
I
EXAMPLE 17-2
Identical types of bearings have a Weibull times-to-failure distribu
tion with the following parameters:
0 = 2, tj = 2,000hr, 7 = 0 hr.
1
Determine the following:
1. If 100 of such bearings are operating in identical equipment at
the same application and operation stress level, how many will
survive if each one operates T\ = 1,300 hr at 675 rpm, and how
many will fail?
2. If the failed bearings are replaced by new ones and all of them
operate for T = 700 hr thereafter, how many of the fresh ones
will fail?
3. Of those that survived T\ hr, how many will fail during the ad
ditional T hr of operation?
4. What is the total number of bearings that will fail by (Ti + T) hr
of operation, given that 100 started at age zero, and that those
that fail by T\ hr are replaced?
5. What is the total number of bearings surviving after {T\ + T)
hr of operation when we follow the policy of replacing the failed
ones by T\ hr?
6. What is the total combined reliability of the components under
the replacement policy?
7. What is the reliability for (T\ + T) hr of operation without a
replacement policy? Compare the answer with that of Case 6.
748 RELIABILITY OF COMPONENTS
,1,300 *2
R( 1,300 hr) = e"( 2,ooo ) - 0.522046.
2. If the failed bearings are replaced by new ones and all of them
operate for T = 700 hr thereafter, the number of fresh ones which
will fail can be determined using Eq. (17.2), or
where
lT\B i 700 i2
R{T= 700 hr) = e"(n» = e~(^^> =0.884706.
Then,
3. Of those that survived T\ hr, the number that will fail during
the additional T hr of operation, from Eq. (17.4), is
R&+T)
NF-NR(TUT) = N0R(T1) 1
Rffl)
From Case 1,
NsiTi) = NoRiTi) = 100(0.522046) = 52.2046, say 52.
and
, Tj+Ts0 ,1,300+700 J
R{Tx + T) = e~[ n i = e~( 2.°°° } = 0.367879.
or
or
Ns-r^ + T)
Rt-r(T, + T) =
N0
79 07
RT-r(1,30Q hr + 700 hr) = #T_fl(2,000) = ——- = 0.7907.
since,
36.8% < 79.1%,
or 5(2, 000 hr) with replacement is more than double that without
replacement.
PROBLEMS
17-1. Derive the equation for calculating the total combined reliability
if the age of the units which have been put into operation at the
beginning of the mission is not zero, or To ^ 0, but T\.
17-2. Find the total combined reliability for the following replacement
policy: At the prescribed replacement time T\, those that are
found to have failed by T\ are replaced by fresh ones, and p
percent of those that have not failed by T\ are also replaced by
fresh ones.
17-3. Find the optimal value of the p given in Problem 17-2 in terms
of cost, assuming that
C\ = replacement cost per unit,
C<i = unit cost,
Cz = salvage value of an unfailed unit at replacement
time T\,
and
d = failure cost at the end of the mission.
3. Of those that survived T\ hr, how many will fail during the
additional T hr of operation?
4. What is the total number of bearings that will fail by (Ti +
T) hr of operation, given that 100 started at age zero, and
that those that fail by T\ hr are replaced?
5. What is the total number of bearings surviving after (T\ +T)
hr of operation when we follow the policy of replacing the
failed ones by T\ hr?
6. What is the total combined reliability of the components
under the replacement policy?
7. What is the reliability for (T\ +T) hr of operation without a
replacement policy? Compare the answer with that of Case
6.
17-8. Using the replacement policy of Problem 17-2 determine the
total combined reliability of the components given in Problem
17-7 assuming that the percent, p, of those that have not failed
by T\ = 1,500 hr, but are replaced, is 30%.
17-9. Identical units have a Weibull times-to-failure distribution with
the following parameters: (3 — 1.75, 77 = 3,000 hr and 7 = 0
hr. The replacement policy is as follows: Those that are found
to have failed at 1,500 hr are replaced with units which have to
have a Weibull times-to-failure distribution with the following
parameters: 0 = 2.50, 77 = 1,500 hr and 7 = 0 hr. Do the
following:
1. If 1,000 such units are operating in identical equipment at
the same application and operation stress level, how many
will fail if each one operates T\ = 1 , 500 hr?
2. If the failed ones are replaced according to the given policy,
how many will fail when operating for t = 500 hr thereafter?
3. Of those that survive T\ = 1,500 hr, how many will fail
during the additional t = 500 hr of operation?
4. What is the total combined reliability of these components
under the replacement policy of Case 3?
17-10. Using the replacement policy of Problem 17-2 determine the
total combined reliability of the components given in Problem
17-9 assuming that the percent, p, of those that have not failed
by ni = 500,000 cycles but are replaced, is 30%.
Chapter 18
RELIABILITY OF MAINTAINED
EQUIPMENT WHEN KEEPING
TRACK OF THE AGES OF ALL
COMPONENTS WITH
EXPONENTIAL AND
WEIBULLIAN PDF'S
where
wo.^^,-^, (18.2)
755
756 RELIABILITY OF MAINTAINED EQUIPMENT
Nci
-r=- = proportion of identical components which fail due
1 to chance causes,
N i
-r— = proportion of the same identical components
1 which fail due to wear-out causes,
N ■
•"ci N ■
, lywx i
Ni + Ni ~ '
anc
N{ = total number of components in the equipment, assum
ing the components exhibit chance as well as wear-out
failures, as would be the general case.
The parameters iVc,-, Nw{, A,-, 7,-, 77; and /?,- in Eq. (18.2) may be de
termined using known techniques [1, pp. 215-263; 271-331].
After the first mission, the reliability of this equipment for the
second mission is given by
N
R2s(Tut) = J[R2i(T2,t),
t=i
where T2 is the age at the beginning of the second mission and t is the
duration of the second mission, or
Rjs(Tj,t)>RGOAL(t)-
MEAN LIFE 757
where the T, are the ages of the respective components and all parame
ters are for these ages, and Tj is the age of the system at the beginning
of the jth mission.
1. Find the system's reliability for the third mission of 50 hr, if the
duration of the first and second missions was 35 hr and they both
were successful.
2. If Unit 1 fails at 40 hr during the first 50-hr mission and then
Unit 3 fails at 30 hr during the second 50-hr mission, find the
system's reliability for the third 50-hr mission.
3. Determine the mean fife of the system in Case 2, for the first
three missions.
4. Discuss the results in Cases 1 and 2 comparatively.
The units that fail during a mission are replaced with fresh ones at the
beginning of the next mission.
TABLE 18.1 - The parameters of the times-to- failure dis
tribution of each unit in Example 18—1.
1 2 3 4 5 6 7
Number
of A,-, 7»i m,
unit * * fr/106 hr hr hr ft
1 0.30 0.70 50 10 400 1.8
2 0.25 0.75 45 50 450 1.5
3 0.20 0.80 70 0 500 2.0
2. From Eq. (18.5) the reliability of the system, keeping track of the
age of each unit, is given by
70-50 ■,, 5
0.25 e-°-000045x70 + 0.75 e^Tsr)
/ 50-0^2.0
0.20 e-0000070x50 + 0.80 e V 500 I
i0-0yi n
0.20 e-0000070x0 + 0.80 e-(w
<■ 500 '
3. The system's mean life for the first mission, is given by Eq. (18.6),
or
Mxs = r Ris(T) dT,
Jo
and from Eq. (18.2)
Jroo »
'Nc
'ii »-A,T i Nw' „-(
M \s + dT,
Ni Ni
or
Mi5=/ {[0.30e-OOOOO5or + 0.70]
•[0.25e-OOOOO45T + 0.75]
.[0.20 e-0000070 T + 0.80 e-(w)2°]} dr
+ y5° { [0.30 e'0000050 T + 0.70 e~<W]
•[0.25e-OOOOO45T + 0.75]
• [0.20 e-0000070 T + 0.80 e-(w)2°]} dT
+ f* { [0.30 e-0000050 T + 0.70 e"<W]
,-0.000045 T
J/-oo 3
My. Ifr e-A,r. + J^ e-(2^)/3• dt, (18.13)
From these results, it may be seen that the system's mean life
for the first mission is the longest. This is always true, because
all units in the system are fresh for the first mission.
4. In Cases 1 and 2, the cumulative mission times are the same and
are equal to 120 hr, but the reliability values are different. The
reliability of the system in Case 2 is greater than that in Case 1;
i.e.,
[iistCase 1) = 89.08%] < [i?s(Case 2) = 92.71%].
The reason is that in Case 2, there are two fresh units which are
the replacements for the two units that failed during the first two
missions.
PROBLEMS
1. Find the mean life to the first failure of the machine, or when
each failed unit is replaced by an identical one that had op
erated a period equal to the period of operation of the failed
unit that is being replaced by a fresh unit.
2. Find the mean time between the second and third failures if
Unit 1 fails first at the age of 500 hr and is replaced with a
new identical unit, and Unit 3 fails next at the age of 7,000
hr and is replaced with a new identical unit.
3. Find the stabilized mean time between failures of this ma
chine, when the failed units are replaced by fresh units.
4. Calculate and plot the reliabilities for the follow
ing two cases:
4.1 Case 1.
4.2 Case 2.
18-2. In a three-unit system, the units function reliabilitywise in series.
Each unit can fail due to chance as well as wear-out causes. The
parameters of each unit's times-to-failure distribution are listed
in Table 18.3. Assume that only the failed unit is replaced and
then the system is readied for the next mission. Do the following:
1. Find the system's reliability for the third mission of 50 hr, if
the duration of the first and second missions was 35 hr and
both missions were successful.
2. If Unit 1 fails at 40 hr during the first 50-hr mission and then
Unit 3 fails at 30 hr during the second 50-hr mission, find the
system's reliability for the third 50-hr mission.
3. Determine the mean life of the system in Case 2, for the first
three missions.
PROBLEMS 763
Number
of A,, 7ii Wi
unit * N fr/106 hr hr hr ft
1 0.15 0.85 40 20 500 2.0
2 0.20 0.80 45 50 300 1.5
3 0.10 0.90 70 0 400 2.5
2. Find the mean time between the second and third failures if
Unit 1 fails first at the age of 600 hr and is replaced with a
new identical unit, and Unit 3 fails next at the age of 5,000
hr and is replaced with a new identical unit.
3. Find the stabilized mean time between failures of this ma
chine, when the failed units are replaced by fresh units.
4. Calculate and plot the reliabilities for the following two cases:
4.1 Case 1.
4.2 Case 2.
18-5. Work out Problem 18-1 again taking ^ = 0.2 and ^p = 0.8 and
compare the results with those obtained in Problem 18-1.
18-6. Work out Problem 18-2 with the parameters of each unit's times-
to-failure distribution given in Table 18.6. Compare the results
with those obtained in Problem 18-2.
18-7. In a three-unit system, the units function reliabilitywise in series.
Each unit can fail due to chance as well as wear-out causes. The
parameters of each unit's times-to-failure distribution are listed
in Table 18.7. Assume that only the failed unit is replaced and
then the system is readied for the next mission. Do the following:
1. Find the system's reliability for the fourth mission of 50 hr, if
the duration of the first, second and third missions was 50 hr
and all missions were successful.
2. If Unit 1 fails at 30 hr during the first 50-hr mission, then Unit
3 fails at 30 hr during the second 50-hr mission, and Unit 1
fails at 40 hr during the third 50-hr mission, find the system's
reliability for the fourth 50-hr mission.
766 PROBLEMS
3. Determine the mean life of the system in Case 2, for the first
four missions.
4. Discuss the results in Cases 1 and 2 comparatively.
The units that fail during a mission are replaced with fresh ones
at the beginning of the next mission.
1. Find the mean life to the first failure of the machine, or when
each failed unit is replaced by an identical one that had op
erated a period equal to the period of operation of the failed
unit that is being replaced by a fresh unit.
PROBLEMS 767
2. Find the mean time between the third and fourth failures if
Unit 1 fails first at the age of 500 hr and is replaced with a
new identical unit, Unit 3 fails next at the age of 7,000 hr and
is replaced with a new identical unit, and Unit 4 fails next at
the age of 8,000 hr and is replaced with a new identical unit.
3. Find the stabilized mean time between failures of this ma
chine, when the failed units are replaced by fresh units.
18-9. Work out Problem 18-8 again assuming that jf = 0.3 and ^ff- =
0.7 for all three units and compare the results with those obtained
in Problem 18-8.
18-10. In a three-unit system, the units function reliabilitywise in se
ries. Each unit can fail due to chance as well as wear-out causes.
The parameters of each unit's times-to-failure distribution are
listed in Table 18.9. Assume that only the failed unit is replaced
and then the system is readied for the next mission. Do the
following:
1 . Find the system's reliability for the third mission of 50 hr, if
the duration of the first and second missions was 35 hr and
they both were successful.
2. If Unit 1 fails at 40 hr during the first 50-hr mission and then
Unit 3 fails at 30 hr during the second 50-hr mission, find the
system's reliability for the third 50-hr mission.
3. Determine the mean life of the system in Case 2, for the first
three missions.
4. Discuss the results in Cases 1 and 2 comparatively.
The units that fail during a mission are replaced with fresh ones
at the beginning of the next mission.
768 REFERENCE
REFERENCE
769
770 INDEX
for units that fail by a pre Steady State Mean Times, 123-
scribed operating time, 139
546 active repair, 123
inventory costs consideration, corrective maintenance, 138
609 corrective restoration, 123
availability maximization preventive maintenance, 138,
per cost ratio model, 615 243-310
dynamic spares provision Summary of developments of main
ing approach, 612 tenance policies, 446
understock and overstock System Effectiveness, 21-36
cost model, 609 concepts of, 24-25
under preventive maintenance definitions, 24-25
policies, 551 quantification, 27-30
age replacement policy, 551 System moments method, 639
age replacement with min repair time distribution de
imal repair, 560 termination, 638
block replacement policy,
554
group replacement policy, Tapco case study, 11
557 Taylor series expansion, 534
modified block replacement for Poisson pdf explanation,
policy, 564 534
modified periodic replace Theoretical considerations for op
ment policy, 559 timum spare kit deter
optional replacement pol mination, 591
icy, 568 Time categories
ordinary periodic replace active repair time, 25
ment policy, 559 administrative time, 25, 41
with cost consideration, 583 down time, 25
Kettelle algorithm, 598 free time, 25
optimum spares kit, 583 logistics time, 25, 41
optimum spares in a kit, operating time, 25
583 storage time, 25
theoretical considerations, times associated with main
591 tenance, 43-44
with growth and warranty,
U
573
Spares Requirement for Policy Understock cost model for spares
1,273 provisioning, 609
776 INDEX
Warranty, 573
spares provisioning, 573
Weibull distribution, 111-114
Weibull process
composite Weibull process
721-724,
graphical estimates, 701-709
maximum likelihood estimates,
709-711
when testing more than one
unit, 711-721
WSEIAC reports, 21-22
ABOUT THE AUTHOR
Dr. Dimitri B. Kececioglu, P.E., a Fullbright Scholar, a Fellow
of the Society of Automotive Engineers, and a Fellow of the Ameri
can Society for Quality, is considered to be the Deming of Reliability
Engineering. He received his B.S.M.E. from Robert College, Istanbul,
Turkey in 1942, and his M.S. in Industrial Engineering in 1948 and his
Ph.D. in Engineering Mechanics in 1953, both from Purdue Univer
sity, Lafayette, Indiana. He is currently a Professor in the Department
of Aerospace and Mechanical Engineering, The University of Arizona;
Professor-in-Charge of a unique ten-course Reliability Engineering pro
gram leading to the Master of Science degree in the Reliability Engi
neering Option; Director of the Annual Reliability Engineering and
Management Institute; Director of the Annual Reliability Testing In
stitute; Director of the Applied Reliability Engineering and Product
Assurance Institute for Engineers and Managers; and a Reliability and
Maintainability Engineering consultant.
This book is based on the following extensive experience of the
author in Reliability Engineering, Maintainability Engineering, Reli
ability & Life Testing, Mechanical Reliability, Environmental Stress
Screening, and Burn-In Testing:
10. He has been consulted extensively by over 100 industries and gov
ernment agencies worldwide on Reliability Engineering, Reliabil
ity & Life Testing, Maintainability Engineering, and Mechanical
Reliability matters.
17. He was elected to the presigious Fellow Member grade of the Soci
ety of Automotive Engineers International in 1996 for "his excep
tional professional distinction and important technical achieve
ments" .