Chapter 4

RELIABILITY BASED SPARE

PARTS FORECASTING AND
PROCUREMENT STRATEGIES

A. K. Sheikh, M. Younas
Department of Mechanical Engineering
King Fahd University of Petroleum and Minerals
Dhahran 31261, Saudi Arabia

and

A. Raouf
Department of Systems Engineering
King Fahd University of Petroleum and Minerals
Dhahran 31261, Saudi Arabia

Abstract Spare parts procurement strategies based on the maintenance engineering

techniques of reliability engineering are merged with materials manage-
ment discipline to provide a practical method to manage and control spare
parts for industry. The well-known techniques of reliability engineering
are used to determine failure rates for equipment and related parts. Then
this information from the maintenance discipline is linked to the data of
the materials management discipline. The results of this work will pro-
vide a scientific method of spare parts forecasting based on reliability of
the parts, and more rational inventory management and procurement
strategies with a minimum risk of stock out. As a consequence overstock-
ing can be eliminated, and spare parts management can be streamlined
on a rationale basis.

Keywords: Spare Parts, Nonrepairable Parts, Reliability, Failure Rates, Renewal

Analysis , Loss Matrix , Inventory Management , Rational Inventory ,
Ordering Strategies .

M. Ben-Daya et al. (eds.), Maintenance, Modeling and Optimization

© Springer Science+Business Media New York 2000
82 MAINTENANCE, MODELING AND OPTIMIZATION

1. INTRODUCTION
With the expansion of high technology equipment in industries world-
wide the need for spare parts to maximize the utilization of this equip-
ment is paramount. Sound spare parts management improves produc-
tivity by reducing idle machine time and increasing resource utilization
[22]. It is obvious that spares provisioning [3, 17, 22, 23, 24] is a complex
problem and requires an accurate analysis of all conditions and factors
that affect the selection of appropriate spare provisioning models. In the
literature there are large numbers of papers in the general area of spare
provisioning. Most of these papers deal with the repairable systems, and
provide a queuing theory approach to determine spare parts stock on
hand to ensure a specified availability of the system.
These queuing theory based models of repairable systems provide an
important contribution to spares provisioning and help to design ade-
quate repair facilities to meet an availability target. These models [1, 4,
8, 9] have been further extended to incorporate the inventory manage-
ment aspect of maintenance [11, 13, 14, 15, 25, 27, 28]. The common
features of these models presented in literature are:

1. They all deal with a repairable system, where in addition to MTTF,

it is necessary to have some knowledge of MTTR . The criterion
for evaluating such a system is its instantaneous availability A(t)
or asymptotic availability, A(oo)= MTTF/(MTTF+MTTR).
2. Based on a queuing theory structure with a demand rate A, and
repair rate J.L , appropriate analytical models are developed. In
some cases demand rate is treated interchangeably with failure
rate. There is a catch in this, since the failure rate is based on
operational time to failure, where the demand rate (used in in-
ventory models) and repair rate (used in availability models ) are
based on calendar time. Since most of the papers do not make this
distinction clear, therefore the models or their parameters need to
be modified to bring all quantities into a calendar time domain.
3. These queuing theory based models primarily deal with constant
failure rates, and constant repair rates. This implies exponential
time to failure and time to repair models. As a process the time to
failure and time to completion of the repair are being character-
ized by a simple homogeneous Poisson process. This assumption
is very restrictive. Of course there are a number of parts such
as electronic components which have a constant failure rate, but
there are hundreds of other mechanical parts which do not con-
form to the exponential time to failure distribution (i.e., constant
 Reliability Based Spare Parts Forecasting and Procurement Strategies 83

failure rate), F(t) = 1- e->'t. These mechanical parts fail due to

aging with time. The aging or wear out mechanisms such as creep,
fatigue, corrosion, oxidation, diffusion, and wear are all time de-
pendent processes, which result in increasing failure rates of the
parts characterized by the Weibull model, F(t) = 1- e-(t/ry){3 with
(3 > 1. Thus the models presented in the literature, which are
for (3 = 1, cannot be used when (3 > 1, because the time to fail-
ure process will have to be treated as a non-homogeneous Poisson
process.

The effectiveness of spare parts management is based on factors which

require improvements in data acquisition and methods of forecasting the
spare parts requirements, analyzing the data on demand of such parts,
and developing proper stocking and ordering criteria for these parts.
The process to obtain meaningful data begins with part identification
and usage information. Usually parts can be classified as unique or com-
mon, critical or non-critical to the operation or equipment. From this
classification the process of data collection can begin. The information
required usually includes lead-time, usage rates, and population of par-
ent equipment from which a forecast of spare parts stocking levels is
determined. However, this type of data is usually inadequate to optimize
the maximum/minimum quantity of spare parts to be stocked based on
some "guess". Similarly forecast of data using time series analysis or
other forecasting techniques [6, 12] using part utilization rate (as deter-
mined by the history of issuing of these parts from store) in general will
not reflect the real situation because of the intermittent use of several
of these parts. As a consequence maintenance departments normally re-
quest more parts than optimum because they have no responsibility for
inventory investments. In order to be on the safe side more parts than
necessary are often requested particularly in huge organizations. Mate-
rial managers have to gain the assistance of the maintenance engineer to
provide the failure analysis and part reliability data in order to improve
spare parts availability. Once the parts or equipment reliability data is
available to the maintenance engineer, both the buyer and the engineer
will achieve mutually beneficial results. Quantitative techniques based
on reliability theory [10, 16, 18, 19, 21] need to be used for developing
the failure rates of the required parts to be purchased and/or stocked.
This failure rate could be used to determine more accurate demand rates.
Once the parts have been identified, for purchase or stock, the deter-
mination of the quantities is required. The quantity, based on failure
analysis, is dependent on the population of the equipment using the
part(s). A reasonable estimate of the failure rate of the part can be
84 MAINTENANCE, MODELING AND OPTIMIZATION

based on historical time to failure data from data points as small as five.
In the absence of such data, the expected failure rates can be obtained
from the information supplied by the manufacturers, which can be cou-
pled with other information from failure rate data banks. This can be
used as an initial starting point, which can be corrected progressively
as more real life in-plant data is available. It is unlikely that sufficient
past history of demand exists for spare parts having irregular or lumpy
demand, or for new parts, making the need for an estimate of the de-
mand based on failure rate and reliability analysis such as the Wei bull
analysis necessary.
Once the usage data and its characteristics have been obtained, the
next step is to develop appropriate inventory and order policy consistent
with an acceptable risk of shortage (tolerable stock-out penalty), inven-
tory investment, criticality of the system, availability (lead time of the
part) and location of the industry with respect to supply sources. After
making the forecast for a specified planning horizon for example say 2 to
5 years, it is necessary to develop a purchasing/procurement policy. A
good parts procurement or ordering strategy must carefully balance the
conflicting cost elements involved in inventory. Ideally one should follow
the economic order quantity equations using the probabilistic inventory
model [31].
The choice is not always so simple; there are other important factors
such as cost and criticality of the part, which influence the decision about
how much to order and when to order. Therefore, spare parts need to
be evaluated in terms of cost and criticality [29]. The cost relates to
purchase cost and is classified as low, moderate or high. The cost of not
having a part which shuts down the system or process is one extreme.
The other extreme is that alternatives are readily available so that a part
loss will not stop the continuation of the process, for example, redundant
systems are in place.

2. SCOPE OF THE WORK

As mentioned earlier, in the broader area of spare parts management
for repairable facilities, a lot of work has been done and published in OR
and Management Journals. Moreover in the specific area of spare parts
management of nonrepairable systems which often fail with time depen-
dent failure rates (/3 > 1), there are some renewal theory based predic-
tion models available for forecasting the needs of spares in a planning
horizon [10]. However, the link between these forecasts and inventory
management (ordering strategies ) is missing.
 Reliability Based Spare Parts Forecasting and Procurement Strategies 85

Increasingly the technology is developing in such a way that for several

types of spare parts , subassemblies and modules, replacing them upon
failure is more economical than repairing them. There are hundreds of
such subsystems and parts in any major organization, representing an
inventory of millions of dollars. Bearings, seals, gears, probes, tubes,
electronic modules, computer parts, gaskets, filters, airplane tires, light
bulbs, turbine blades, valves are some of such parts which are mostly
replaced rather than repaired. In a case where some of these parts fail
due to a minor repairable malfunction, then such a part will be replaced
with a new one while the malfunctioning part is send to a workshop
or other repair facility. After repair this repaired part is reinstalled
in the system of the next opportunity. The time to malfunction will
not be treated as a time to failure of the part, and will not be used
to determine failure rate (and demand rate). After the parthas been
reinstalled the additional time to failure will be added in the previous
time to malfunction of this part, and the resulting total time will be
used to determine failure rate. The time taken in dismantling a part
and installing a new part has been considered to be negligible in this
paper, because for reliability characterization this time is not relevant.
This is a part of MTTR and is only needed if the availability analysis is to
be performed. The focus of the present work is to develop rational spare
forecast and inventory management strategies for non-repairable parts
or systems. There is no restriction on whether parts fail at a constant
rate or an increasing rate. The failure process will be characterized as
a renewal process with instantaneous replacements. After obtaining a
forecast for the planning horizon, ordering strategies are proposed for
different categories of parts in view of their cost and criticality to the
operation. Since demand rate for the spares is derived from actual part
failure rate, the decisions are truly reliability based.
A scientific approach to spare parts forecasting and management
along these lines was proposed in two earlier papers [29, 30]. Based
on the ideas and concepts presented in these papers a formal method
of determining spare parts needed in a given planning horizon, and
their stocking levels, and ordering strategies , is provided with sufficient
details and illustrative examples. Less formal methods of spare part
inventory policy have not always worked particularly in manufacturing
situations. What is needed is a well planned, integrated methodology
beginning with equipment population, and the reliability characteris-
tics such as average life, coefficient of variation, Wei bull parameter and
failure rates . Then one needs to link these aspects of a maintenance
system with the terminology of inventory control (Le. Economic Order
Quantity (EOQ) and demand rate). Such an approach is developed to
86 MAINTENANCE, MODELING AND OPTIMIZATION

accurately forecast the spare parts requirements and to provide rational

part ordering strategies, which will result in minimum inventory with a
higher level of confidence that there is a reduced risk of parts being out
of stock when needed.

3. RELIABILITY CHARACTERISTICS OF
THE UNIT
Reliability defines how long equipment will run without a failure.
It defines the probability of the part not failing within the specified
time. Since reliability is important to the maintenance effort, it must be
included in spare part forecasting techniques. Buyers should recognize
that parts have varying degrees of reliability and failure rates . Based
on the assessment of reliability and failure rates, buyers can develop
better spare parts management strategies.
Reliability is the mathematical probability that a product will func-
tion for a stipulated period of time 't' and can be expressed as a Relia-
bility Function R(t) of the part as follows:
00
R(t) = P(T > t) = J f(t)dt (4.1)
t
where, f(t) is the probability density function of the time to failure of
the part. T is the random variable life whereas 't' represents its values.
Another useful function to represent the failure pattern of the parts is
the" hazard function" or "failure rate function" ,X( t), which is defined as

P[t < T < t + ~t] f(t)f).t

'x(t)f).t = P [t < T < t + f).t IT>t] = P[T> t] = R(t)
(4.2)
or
,X(t) = f(t)/ R(t) (4.3)

3.1 Weibull Reliability Model. The Wei bull reliability

model is a most versatile model for characterizing the life of machine
parts. It is a two parameter model given by the following reliability and
related functions:

R(t) = exp[-(t/17 ).6] (4.4)

f(t) = (f3/17) (t/17 ).6- 1 exp[-(t/17 ).6] (4.5)

'x(t) = (f3I11) (t/17 ).6- 1 (4.6)

 Reliability Based Spare Parts Forecasting and Procurement Strategies 87

where t > 0, TJ > 0 and /3 > O. The parameter TJ is the" characteristics

life" parameter. It has the same units as t, and the reliability at t = TJ
is R(TJ) = 0.368. The mean J-l and variance (]"2 are given by the following
equations:

J-l = TJr (1 + 1//3) (4.7)

(]"2 = TJ2 (r (1 + 2//3) - r2 (1 + 1//3) ) (4.8)

The parameter /3 is a shape parameter and is a non-dimensional quantity
only dependent upon the coefficient of variation of the time to failure of
the part, i.e., /3 = cp((]"/J-l) (Figure 4.1).
The versatility of the Wei bull model is demonstrated in Figures 4.2
and 4.3 where failure rate functions >.(t), and probability density func-
tions for /3 > 1, /3 = 1, and /3 < 1 are plotted. In service either the
components fail randomly (i.e., during their useful life period) which
is reflected by /3 = 1, i.e., the constant failure rate, or they wear out
with the passage of time (/3 > 1) representing an increasing failure rate.
Therefore, this paper will primarily focus on these two cases of the fail-
ure rate. Determination of shape parameter and characteristics life TJ
can either be made by analyzing the historical time to failure data, or, in
the case of absence of such data, one can refer to the values tabulated in
Ref. [5] as starting values. A condensed table (Table 4.1) derived from
Ref. [5] is also given in this paper. The initial assessment about TJ can be
made by using the average life T = J-l as provided by the manufacturer of
the part, or determined from the past experience about the average life
of similar parts. These estimates are then progressively updated when
in house part failure history is accumulated.

3.2 Determining Parameters f3 and 'r/ From Time

To Failure Data. Equation (4.4) can be transformed as Inln(l/ R(t))
=/3lnt - /3lnTJ ,which is an equation of a straight line. If historical time
to failure data tl < t2 < t3 < < ti < tN is available, then the relia-
bility function R(ti) can be estimated using the well known mean rank
formula [18]:

R(t.) = N +1- i (4.9)

~ N +1
88 MAINTENANCE, MODELING AND OPTIMIZATION

~
0
II
~
10°
~
........ "
Ef
~ "-
'5 r--...
....>0 I'---~
.......
r-- t---r-..
i
'0 10- 1 r-.....
Sq,l
0
U

2 4 20

Figure 4.1 The relationship between the coefficient of variation K = (j / J.L and shape
parameter ,,8 of the Weibull model [,8 = 1/K ).

Table 4.1 Failure Modes and Weibull Shape Parameter ,8 [5).

I Failure Mode I Weibull Shape Parameter j3 I

1. Deformation 1.0
2. Fracture 1.0
3. Change of Material Quality
(a) Ageing 3.0
(b) Degradation 2.0
(c) Burning 1.0
(d) Embrittlement 1.0
4. Corrosion 2.0
5. Wear 3.0
6. Leakage 1.5
 Reliability Based Spare Parts Forecasting and Procurement Strategies 89

l(t)

~>1

~= 1

Op erating Time, t

Figure 4.2 Failure rate function of Weibull model for f3 = 1, f3 < 1 and f3 > 1.

ft:t)

Operating Time, t

Figure 4.3 Probability density function of Weibull model for f3 = l(constant failure
rate), f3 < 1 (decreasing failure rate) and f3 > 1 (increasing failure rate).
90 MAINTENANCE, MODELING AND OPTIMIZATION

and lnln(l/ R(ti)) can be plotted against lnti , and a straight line can be
fitted to this data by the linear regression approach. The parameters of
the fitted line (slope and intercept) can be used to determine the Weibull
parameters {3 and 'T} as explained in Ref. [28].
Example 1
For illustrative purposes time to failure data of 54 nominally identi-
cal extrusion dies used in an Aluminum extrusion plant is analyzed and
results are plotted in Figure 4.4. The time to failure of a die is propor-
tional to the weight of metal being extruded until it fails. Thus the life
in this particular situation is represented by kg of metal being extruded.
High coefficient of determination R2 = .995 indicates an excellent fit to
the data.
The shape and scale parameters of these dies are {3 = 1.83 and 'T} =
25730 kg.
Since {3 > 1, this represents a wear out mode of failure. Other impor-
tant reliability indicators of these dies are:

JL 'T}r (1 + 1/(3) = 25730r ((1 + 1/1.83)) = 22850 kg

(T 'T} (r (1 + 2/(3) - r 2 (1 + 1/(3))0.5

25730 (r (1 + 2/1.83) - r 2 (1 + 1/1.83)) 0.5 = 11073 kg

Coefficient of Variation of die life can be found from Figure 4.1 as 0.57,
or can be determined approximately as K = 1/{3 = 1/1.83 = 0.546.

4. SPARE PARTS CALCULATIONS FOR

FAILURE REPLACEMENT
4.1 Failure Replacements. Replacements of individual units
are made just after their failure. This type of replacement is primarily
made for parts which will not cause any further damage to the system
(machine) due to their failure in operation. If the parts are such that
their actual failure in service may result in damage to the other parts of
the system, then condition monitoring [20] is necessary and the replace-
ment can be made just prior to the failure or somewhat earlier than that.
In other words the time to replace a part under the condition monitoring
scheme is when a certain manifestation of progressive damage reaches a
critical level.

4.2 Failure Replacements Process Characteriza-

tion by Renewal Function. One realization of the renewal or
counting process of failure replacements is given in Figure 4.5.
 Reliability Based Spare Parts Forecasting and Procurement Strategies 91

2.00 .....................,..........

·········r·········r·········r·········r·········r·········r·········:··········:·········
I , , ,

1.00
• I , ,

I I
,

I
,
,

,
,
I

I
,
,

, I

0.00 - - - - - - - - - ~- - - - - - - - - - ~ - - - - - - - - - ~- - - - - - - - - -~ - - - - - - - - - ~- - - - - - - - - - ~ - - - - - - - - _.J
, ,
,, ,,
, , ,
I I I I ,
~ -1.00 ---------,----------r---------"'I----------r---------,
I I I 1
I , , I

,
j
,
, I , I

, I I I , , I I
-2.00 --- --- --- .. ------. ---~-- ______ - .. ---- - -,.- -------- .. -- -------- .. -- --- ----1- __ .-- -- - -,.---- --- -- 1-- --_.

,
I I , , ,
,
_ _____ ___ • .1 __________ 1.. ________ ..1 __________ 1. _______ ._..1. ________ .1.I _________ ..1I _____ •
-3.00
~

I I I I , , I I
, I I , " ,

,, "
, r" I
I

, I t I

--------1----------r---------1----------r---------1--- -------r---------l----------r---------l------
, " " I
-4.00
, , , I I I I ,
I , I , ,
I I I I ,

-5.00 -1-----;.--..;-----;---:-'--.;...'--.;...'--...;.'--~'--_:_-
B.OO B.30 B.60 B.90 9.20 9.50 9.BO 10.10 10.40 10.70

lnlKI

Figure 4,4 Weibull reliability analysis of 54 extrusion eies life :die life Xi is expressed
as Kg of metal extruded until failure.

Failure Replacement

tj t2 t3 4 ts i6
t.. ..JI ..JI ..JI
. . t..
..JI

Figure 4.5 One realization of renewal or counting process of part replacement

92 MAINTENANCE, MODELING AND OPTIMIZATION

The renewal function H(t) = E[N(t)] = N(t) defines the expected

number of failures in time t and is give by the following integral equation
[22]
t
H(t) = E[N(t)] = F(t) + J H(t - x)dF(x). (4.10)
o

The renewal rate function h(t) = dH(t)/dt gives the expected number
of renewals per unit time, and satisfies the following integral equation
[10]:
t
h(t) = dE[N(t)Jldt = J(t) + J h(t - x)J(x)dx. (4.11)
o

The variance of number of renewals V(N(t)) is given by [22, 24]

t
V[N(t)] = 0"2[N(t)] = 2 J H(t - x)dH(x) + H(t) - H2(t) (4.12)
o

Equations (4.10-4.12) are general equation valid for any probability

model. However, by substituting the Weibull cumulative distribution
function for time to failure, F(t) = 1 - R(t), as

F(t) = 1 - exp[-(t/rJ ),8]. (4.13)

these functions can be evaluated for the Weibull Model. The renewal
functions for the Wei bull model for different values of j3 (or coefficient
of variation K = 1/ j3) are given in Figures 4.6-4.8 [32].

4.3 Asymptotic Behavior of Renewal Functions.

Consider replacements of a part having an average time between failure
as T and standard deviation of time between failures as O"(T) (coefficient
of variation of time between failures, K = O"(T)/T) . If the operation
time t of the system or machine on which this part is installed is quite
long and several replacements need to be made during this period, then
the average number of Failures E[N(t)] = H(t) will stabilize to the
following asymptotic value of the renewal Junction [10],

- t I t 1 1
N(t) = H(t) = E[N(t)] = T + "2(K 2 - 1) = T + "2(j32 - 1) (4.14)

and the Jailure intensity or renewal rate Junction is given by

 Reliability Based Spare Parts Forecasting and Procurement Strategies 93

2.0 --.------------------"......--;;=-:::;00:1

If- 1.5
::<:,-

;:::.
I
cO
0
""c<-> 1.0
.2
<Ii
s:
(J)
c
(J)
0:: 0.5

0.0 0.4 0.8 1.2 1.6 2.0 2.4

tIT

Figure 4_6 Renewal function of Weibull model.

h(t) = dH(t) = dE[N(t)] = 1 (4.15)

dt dt T
The variance of number of failures in time t, is

V[N(t)] ~ K2 (~) ~ ;2 (~) (4.16)

The standard deviation of number of failures in time t, is

u[N(t)] = u~) ~ = K~ = (~) ~ (4.17)

Exponential model of time to failure is a special case, having K =

1/(3 = 1, and where 1/"., is the failure rate of the part. Thus for an
exponential model of time between failures, the equations (4.14) and
(4.17) will reduce to equations (4.18) and (4.19) respectively.
94 MAINTENANCE, MODELING AND OPTIMIZATION

10

8
k = 0.050
II-
;:£
.;...; ~
::c 6
~(f)
c
a>
-0
OJ 4
;:
a>
c
a>
(Y

0
00 04 0.8 1.2 1.6 2.0 2.4

tff

Figure 4.7 Renewal rate function of Weibull model.

0.3
Ii='
x
~

6
n->a>- 0.2

0
c
OJ
·c
OJ
>
0.1

00 04 0.8 1.2 1.6 2.0 24

tff

Figure 4.8 Variance of number of renewals for Weibull model.

 Reliability Based Spare Parts Forecasting and Procurement Strategies 95

t
H(t) = E[N(t)J = T (4.18)

o-[N(t)J = # (4.19)

and equation (4.15) will remain unchanged.

4.4 Probability Distribution of N(t) when t is Large.

If time t in equation (4.14-4.19) representing a planning horizon is large,
then from the central limit theorem N(t) is normally distributed with
mean = N(t) and standard deviation o-(N(t)) [22J and is given by:

(4.20)

Then the number of spares N needed during this period with a prob-
ability of shortage = 1 - p, is given by

N = ~ + ~(K2 - 1) + K !+cp-1(p)
(4.21)
= ~ + ~(:b - 1) + ~ff.CP-1(p)
where cp-1(p) is the inverse normal function and is available in proba-
bility textbooks [18],as well as in most of the spreadsheet software such
as Excel. For an exponential model (K = 1) equation (4.21) will reduce
to

N = =t
T
+ #-1
=cp
T
(p) (4.22)

Example 2:
A work piece needs to be manufactured on a machining center. The
milling cutter required for machining has an average life T of 160 min-
utes of cutting time at the adopted machining conditions. The quantity
of jobs demands a total of 26 hours of machining time. With a confi-
dence of 95%, calculate the number of milling cutters needed, (3 = 3 for
the cutter.

Solution: cp-1(p) = CP-1(0.95) = 1.65

T = 160 minutes
{3=3
t = 26 x 60 = 1560 minutes
Using equation (4.21)
96 MAINTENANCE, MODELING AND OPTIMIZATION

1560
N = 160 + '12 ( 1
32 - 1
)
+ (1'3 J1560)
160 (1.65)

N=I1.02 cutters, say 11 cutters.

5. SPARE PART CALCULATIONS FOR

SCHEDULED REPLACEMENTS
5.1 Scheduled Replacement Strategy. Units or parts
are replaced at scheduled times (for example, during scheduled shut-
down maintenance of the machine) or periodic intervals, i.e., ts, 2t s , 3t s , ... , jts,
etc. However, if any failure of the units occurs during these scheduled
intervals, the additional failure replacement/replacements will also be
made. These additional failure replacements will be rather small in
number provided there is an optimal or near optimal selection of the
scheduled replacement intervals ts.
In a planning horizon of duration 0 - t such that t = Nsts, where Ns
represent the scheduled replacement cycles, we have Ns scheduled re-
placements of the part and NsH(t s ) unscheduled or failure replacements
of the parts. Thus total numbers of parts needed in Ns replacement
cycles are

N = Ns[1 + H(ts)J (4.23)

H(ts) can be read from Figure 4.6, or from the Tables of the Weibull
renewal model such as those given by Lipson & Sheth [19J. Equation
(4.23) implies that the parts are being replaced corresponding with re-
liability level R(ts) = exp[-(t s/17),6J. Note N in equation (4.23) will be
greater than given by equation (4.21). Therefore, the scheduled replace-
ments can only be justified if various cost elements and value of f3( or
K = a /T) warrant it.

5.2 Comparison Of Long Run Expected Cost Of

Replacements Under Scheduled And Failure Replace-
ments. Long run expected cost of replacement per unit part for this
type of replacement policy is given by:

E{C(t)} = [C,+CsH(ts)J/ts (4.24)

where Cs is the cost of a scheduled replacement, and C, is the cost of a
failure replacement.. Therefore, the policy of equations (4.23, 4.24) will
only be superior under certain conditions of f3(or K = afT) and Cs/C,.
This policy can be compared with the Failure Replacement Policy, which
has an expected cost
 Reliability Based Spare Parts Forecasting and Procurement Strategies 97

Cf [H; t)] = C f [; +~ (;2 - 1)] (~) = i+ A( t) ~ i. (4.25)

Thus for superiority of the Scheduled Replacement Policy

Cf [(Cs / Cf) + H(ts)] < Cf (4.26)

ts T

[~;] < [; - H(t s )] (4.27)

where H(ts) will depend upon parameter (3. The assumption is that C f
is the same under both policies.

Example 3: Same as Problem 2, when Scheduled Replacement Pol-

icy is adopted and Cs/Cf = 0.6 , and Ts = ts/T = 0.92,

Solution: Cs/Cf = 0.6, K = 0.333, Ts = ts/T = 0.92,

ts = 0.92(160) = 147 minutes
Ns = \5:~ = 10.61 ~ (l1cutters = N s )

Thus 11 schedules replacements are made in 1560 minutes. Additional

cutters due to unscheduled failures during 11 periods of replacements
= Ns[H(t s )] = NsH(ts/T) = N s [H(0.92)] = 10.61(0.6) = 6.3 -+ 6
cutters (where H(0.92) = 0.6 is read from Figure 4.5). Thus, the total
number of cutters needed for the planning horizon of 1560 minutes of
cutting time = 11 + 6 = 17 cutters.
Although the number of cutters needed is more than calculated in
Example 2, the cost of a premature failure Cf in this case is much greater
than the cost of a scheduled failure Cs, which makes consumptions of
17 cutters more economical than a failure replacement policy with its
associated cost of damage or penalty due to cutter failures.
The general rules for selecting the appropriate replacement policies in
view of the above parameters are discussed in Ref [30], and the results
are summarized below:
1. If (3 = 1, then adopt the Failure Replacement Policy.
2. If (3 > 1, but Cf > Cs, then further investigate the possibility of
adopting Scheduled Replacement Policy. Under certain conditions
of (3 and Cs/Cf the Scheduled Replacement Policy will be better.
If desired the optimal interval for scheduled replacements can t:
be obtained from the method shown in Ref [30].
3. If (3 > 1, but C f < Cs, then adopt the Failure Replacement Policy.
98 MAINTENANCE, MODELING AND OPTIMIZATION

6. PURCHASING AND MAINTENANCE

LINK
Purchasing has to work with the maintenance, reliability , or plant
engineer to correlate the actual time in service of a part when the part
usage or precise demand is not known. If usage history is known, based
on sufficient past demand (typically 5 or more data points), then this can
be used directly in the inventory models. It is unlikely that sufficient,
past history of demand exists for spare parts having irregular or lumpy
demand, or for new parts, making the need for an estimate of the demand
based on failure rate and Weibull analysis necessary. Once the failure
intensity is established there is a need to link the usage rate U(t) based
on calendar time (t c ) of the part usage to the failure intensity of the
part, which is based on actual operating time (t). In order to establish
such a link we define the part utilization p as follows:

t
p=- (4.28)
tc
tc is a convenient time period (often consider as one year). Since tc > t,
the range of p will be from 0 to 1 (0 < p < 1). The usage rate per unit
of calendar time U(t) = U and the average operational time between
failures T is related as follows:

- 1
U=~p (4.29)
T
where p ~ 1 is the utilization ratio (capacity ratio, or duty cycle) of the
part to take care of the idle time of the machine. For nonstop operation
of the part p = 1 resulting in U = j;, typical usage rate for various spare
parts can be developed in house for a given plant.
The intensity of failure (i.e., liT ), usage rate, and the demand rate
D of the part are related to each other as follows [29]:

(4.30)

where Ns(t o) = No = total number of parts which are in operation at

some reference time to ~ O. Thus using equations (4.21), (4.28), and
(4.30), the number of parts (N) needed in a calendar time tc is given as
follows:

(4.31)
 Reliability Based Spare Parts Forecasting and Procurement Strategies 99

6.1 Determining Demand Rate From the Knowl-

edge of Usage Rate. The failure intensity can also be calculated
by a less sophisticated method of using historical usage rates. Usage
rate of a spare part Xj associated with a specific machine 'j' can be
calculated, by using historical information of spare part utilization in
the past, with the help of the following equation;

(4.32)

where
U(t)= Average rate of utilization per part pe~ month (year).
I: Xj = Total net issues for spare parts Xj to be used on machines j.
t = Number of month (year) for which transaction history is available.
Qj = Quantity of machines of type j using the parts Xj.
Nj = Number of parts Xj installed on jth machine.

Once the usage data and factors have been obtained, the next step
is to develop the appropriate inventory and order policy consistent with
risk of shortage (stock out penalty), inventory investment, criticality of
the system availability (lead time of the part) and location of the process
with respect to supply sources.

7. SPARE PART CLASSIFICATION

There are other important factors such as cost and criticality of the
part, which influence the decision about how much to order and when
to order. Therefore, spare parts need to be evaluated in terms of cost
and criticality.
The cost relates to purchase cost and is classified as low, moderate
or high. The cost of not having a part which shuts down the system
or process is one extreme. The other extreme is that alternatives are
readily available so that a part loss will not stop the continuation of the
process. For example, redundant systems are in place.
Criticality is based on the cost of not completing the process or
assigned equipment function i.e. "the mission". Criticality can also be
classified as low, moderate or high. Highly critical parts are those
which are absolutely essential for mission success. Moderately critical
parts are such that if they are out of stock at the time of demand, it
will have only a slight to moderate effect on mission success, whereas
parts of low criticality are not absolutely essential for mission success.
If such parts of low criticality are not available on demand, alternate
parts can be substituted, or in-plant manufacturing of such parts is
100 MAINTENANCE, MODELING AND OPTIMIZATION

Table 4.2 Loss matrix to determine ordering policies.

Cost /Criticality Low Moderate High

1 -+ L M H

Low, L l***
LL l***
LM l***
LH
Moderate, M l**
ML l**
MM l'MH
High, H l**
HL IBM lBH

possible, or they are instantly available in the market. To assess the

criticality of parts a number of factors need to be considered, including
the Pareto analysis [21, 26] of the part/equipment failure to establish
ABC classification, and development of a criticality loss matrix [29].

8. PRACTICAL SPARE PARTS

MANAGEMENT
Considering cost and criticality of parts jointly, one can represent
the loss to the organization due to their absence on demand, in the
form of a loss or regret matrix as given in Table 4.2, where liK = loss
due to the event that a part of cost, i, (i = L, M, H) and criticality, k,
(k = L, M, H) is out of stock when demanded. The ordering strategies
can be decided on the basis of this loss matrix , by grouping various
closely related outcomes together (indicated by symbols, *, **, and * * *
in the loss matrix ).
Ordering Strategy 1:
For parts grouped as class 1, symbol *, set [l H M , l H H , l M H]

Maintain a given quantity of these items in the inventory for com-

pensation of repair and recycle times and procurement lead times, etc,
and order new items on a one for one basis when failure occurs, and a
spare is withdrawn from the inventory. This strategy will minimize the
risks involved due to tying up too much capital and the resultant high
inventory maintenance cost. Initial stock level for such parts can be cal-
culated using equations (4.21) or (4.22) depending upon the coefficient of
variation, K, and incorporating a high value of p. For example, a value
of p = 0.999, will imply that there is a very low probability of stock out
in time interval 0 - t~, where t~ is much less than the mission time, but
greater than [Tc + 3£Tc+ (Manufacturer's lead time + Shipment time)]
 Reliability Based Spare Parts Forecasting and Procurement Strategies 101

i.e. t~ ~ [Tc +30"c+(Manufacturer's lead time + Shipment time)], where

all quantities are expressed in calendar time. As an additional factor
of safety the number of parts calculated using equation (4.21) or (4.22)
corresponding to t = t~, can be increased by one in the initial acquisi-
tion. Later on the orders are placed on a one by one basis.

Ordering Strategy 2:
For parts grouped as Class 2, Symbol **, set [lML' lMM, lHL]

In addition to the purchase cost side of the decision process, the other
critical cost factors in terms of service level (risk) and safety stock need to
be considered. The relevant model for ordering strategy 2 is a standard
inventory model used to calculate safety stock associated with a low
probability of stock out. The formula to calculate the optimum reorder
quantity Q* for this model is [21]

(4.33)

where
D= average demand per year, units/year
Co = cost per order, $ / order
Ch = Annual holding and storage cost per unit of average inventory,
($ /units/year)

This formula is consistent with the probabilistic inventory model illus-

trated in Figures 4.9 and 4.10. The model's variables are mentioned in
Figure 4.9 and Figure 4.10, which illustrate the concept of safety stock,
and associated risk of stock out (dark shaded area of the normal prob-
ability curve in Figure 4.10). Average demand during lead time (LT)
is calculated from equations (4.30) and (4.14),and standard deviation of
demand during lead-time is calculated from equations (4.30) and(4.16)
by replacing t = LT, in these equations.
The effect of high stock out costs will increase the order quantity,
which decreases the frequency of exposure to risk of stock out. If the cost
of stock out C e can be assigned to the number of order cycles (exposures
to a stock out) then the formula for the optimum order quantity can be
modified as

Q*= (4.34)
102 MAINTENANCE, MODELING AND OPTIMIZATION

------- ------ --- --- ------- ,- --- ---------- --- ------- --7
'. Order
.
\
,
Point

Reasonable '.
Maximwn . Safety
Stock
Demand
'.'.
,
"
"

Lead Lead
Tune une
Time

Figure 4.9 Illustration of inventory model and related variables.

Ordering Strategy 3:
For parts grouped as Class 3. Symbol ***, set [lLL' lLM, lLHl

These parts are of low cost and varying degree of criticality. There
could be two possible strategies for this type of parts:

Strategy 3(a)
If the following conditions are simultaneously satisfied:

i) spare parts can be stored for an entire mission duration without

any damage during the storage.
ii) Storage cost for these parts is very small.
iii) There is a danger of non-availability of these parts during the mis-
sion time. This danger could be because of unusual circumstances. This
is not inherent in assessment of part criticality.

Parts are not instantly accessible in the local market. They have to
be ordered sufficiently far ahead of time.
 Reliability Based Spare Parts Forecasting and Procurement Strategies 103

Nonnal Distribution
ofDLT(with 1.1. onda)

---1'----

Probability of
S atoty Sto ck Stockout During
This Ordor Cycle

Figure 4.10 Effect of normal demand during lead time.

Then the strategy for ordering is to order them in one lot of N items
to cover the entire mission period. For a specific risk of stock out say
1- p = 0.05, or p = 0.95, and t representing the mission time, the order
quantity Q = N can be calculated using equations (4.21) or (4.22) de-
pending upon the value of K.

Strategy 3(b)

If one or more of the above conditions are not satisfied then make
several smaller orders during the mission period, which will be identical
to strategy 2, discussed in the previous section.

The ordering strategies discussed above cover both the safety stock
and initial stocking level. However, the costs and risks need to be bal-
anced in order to arrive at an inventory stocking level for specific spare
parts . The following section summarizes the overall decision process in
setting industrial policies.

• Spreadsheet based Weibull analysis of time to failure of parts fre-

quently replaced in plant under consideration provides the Wei bull
104 MAINTENANCE, MODELING AND OPTIMIZATION

parameters of these parts. This analysis also leads to characteri-

zation of the modes of failure.
• Reliability models expressing operational time t, are expressed in
term of calendar time t e , using part utilization ratio, (duty cycle,
or capacity ratio); p = to/teo
• Spreadsheet based forecasting of spare parts using Weibull relia-
bility and renewal models is made for a specified planning horizon
depending upon the part replacement strategy i.e. [10, 16, 19, 21J
- Failure replacement
- Scheduled replacement (constant interval replacement)
• The results are expressed both in terms of operational time and
calendar time. Relevant charts (non dimensional graphs) could be
developed to facilitate the decision making.
• From part failure rates , the demand rate of the parts can be es-
tablished and the statistical characteristics of demand (or demand
rate) can be determined which provide the necessary input along
with various inventory cost parameters in appropriate stochas-
tic inventory control model(s). Optimal inventory management
strategies (EOQ, Lead time, etc) are established.
• Factors determining criticality of parts are established. Pareto
analysis is used along with other relevant criteria such as a criti-
cality loss matrix to determine the criticality of parts.
• Part ordering strategies are streamlined depending upon the out-
come of cost and criticality matrix.

Example 4.

An aluminum extrusion plant has signed a contract to supply for 3

years a certain type of extruded cross section of aluminum, at a rate of
200000 kg per year (uniformly spread over the entire year). The historical
time to failure data of the die used for extruding this cross section is re-
trieved from the company's data files and a Weibull analysis is performed
on this data. The results are plotted in Figure 4.4. The total annual
production of all type of aluminum extrusion in the plant is 1000000 kg.
The calendar year, te is proportional to 1000000 kg. The operational
time during the calendar year, t is proportional t0200000 kg. The uti-
lization ratio p = t/te =200000/1000000=0.2.
From Figure 4.4, the Weibull parameters are f3 = 1.83 and", = 25730
 Reliability Based Spare Parts Forecasting and Procurement Strategies 105

kg. The cost of the die (US $5000 / die) is high, and its criticality (being
vital to the operation) is also high. Thus the criticality - cost loss matrix
evaluation is lHH. Thus ordering strategy 1 is most appropriate in this
case. Die Manufacturer's lead time is 7 weeks, shipment time is 1 week.
Thus total lead time and shipment time is 2 months =1/6 years.
T = 22850 kg.
Tc = T / p = 22850/.2 = 114250 kg -+114250 kg /1000000 kg/ year
= 0.114258 year
3ac = 3a/p = 38970 kg -+ 38970 kg /1000000 kg/ year = 0.0389 year
t~ = 0.114258 + 0.0389 + 1/6 = 0.31989 year
Operational time t* = pt~ = 0.063978 year
-+ 0.06397 year x1000000 kg/ year = 63979 Kg
From Figure 4.6, H(t*) can be read for K < 0.35. For larger values of
K, equation (4.14) is valid. Using Equation (4.14) H(t*) = H(63979) =
2.4624 dies ::::}3 dies. Therefore the ordering policy is to have an initial
inventory of 3+ 1=4 dies and then place an order for one die when ever a
die fails. During the three years operation with 1- p = .9999, the total
number of dies needed are forecasted using equation (4.21) , as 33 dies.

Example 5.

In an existing inventory system the following information is available

for a part of moderate cost and low criticality:

Existing Inventory Policy

Item. Filter assembly. Code: XY - WX - YZA.
On hand inventory: 28, Safety stock = 0,
Unit cost: Co = $128 (moderate)
Holding cost: 15% of $128 = $20
Average Demand Rate: D= 13 parts/year
Planning Horizon: = 1 year
Coefficient of Variation of Life: K = 1

Proposed Inventory Policy

Number of spare parts needed in one calendar year with a risk of

being out of stock =0.05 (i.e., p = 0.95), can be calculated using equation
(4.29):
N = 1 x 13 + 19(1x13)o.S(1.65) = 13 + 6 = 19 parts.
106 MAINTENANCE, MODELING AND OPTIMIZATION

9. EXTENDING THE IDEAS TO

REPAIRABLE SYSTEMS
The present paper deals with nonrepair able systems, but could be fur-
ther extended to repairable systems. Early work in this regard reported
by Cooper [7], utilizes the Erlang's loss formula to estimate the spares
during the constant failure rate region (13 = 1). The formula proposed
by Cooper [7J is given by:
s
A(S,M) = (Ms/M!)/:~:)Mk/k!) (4.35)
k=O
A(S, M) = steady state unavailability of the system when the number
of spares on hand is S and the number of units under repair are M.
A(S, M) = 1- ..4(S, M) = availability of the system.
The number of units under repair depends on the total number of
units in service (N), the repair rate (J-l), and the average lead time (l)
for obtaining spares. The average number of units under repair is the
product N J-ll . Thus, the number of spares S required to obtain a spec-
ified availability level, A*(S, M) = 1 - ..4*(S, M), can be obtained by
solving equation (4.35) for S. The repair rate J-l = C>" is a multiple
of failure rate , >.. , where C is a constant. To facilitate the solution
of equation (4.35), graphs have been developed for different ranges of
N J-ll and availability A( S, M) by AT&T [2J. The ideas presented in our
paper can be integrated into the approach of the papers dealing with re-
pairable systems, by relaxing the constant failure rate condition in model
formation .Models can be developed for the availability of a repairable
system, although the resulting equations would be quite complex. This
paper can become a starting point for such a work. This work using
modest mathematics represents a very important segment of the spare
parts management picture. The approaches to extend this work are as
follows:
• There is not much literature available in spare parts forecasting
and their interface with inventory models for aging parts. The time
to failure process will become nonstationary for 13 > 1 this needs
to be embedded in the queuing model formulations. This can be
done, by using the time to failure process as a renewal process, in
developing the appropriate availability equations. Mathematically
this will be complex but the asymptotic results are expected to be
relatively simple.
• Alternatively MTTF based on Wei bull (13 > 1) distribution can
be approximated as 1/>.. for a queuing theory based model. This
 Reliability Based Spare Parts Forecasting and Procurement Strategies 107

approximation of replacing Weibull ((J' I /-L < 1) by exponential

((J' I /-L = 1) will lead to over estimation of demand rate and over-
stocking will result. If 1 < {3 < 1.5, this approach may be justified
on the ground that overstocking will be less; if {3 > 1.5, the dis-
crepancy will be higher.

10. CONCLUDING REMARKS

This paper provides a computer-oriented methodology to calculate re-
liability centered spare parts requirements for nonrepair able systems,
or nonrepair able parts of a repairable system, and their rational in-
ventory management strategies. These strategies are based upon a
stochastic characterization of the time between failures of various fre-
quently failing parts and equipment. This characterization can be used
to determine the appropriate part failure rate, reliability functions and
the replacement models. Then these models are linked with probabilis-
tic inventory control models through part utilization ratio (duty cycle),
p = tlte. In addition to that for various part failure modes the corre-
sponding {3 values are documented which could be the starting point to
use the methodology if in-plant historical time to failure data is not yet
available.
Having assessed the cost of ordering and the cost of stock out based on
the safety stock analysis, a decision process can be developed to balance
the results and determine a spare parts policy. This process can take
the form of the following decision rules:

1. If purchase cost is low and critical cost is low to high and the part
is readily available then maintain stock at local supplier or use just
in time techniques.

2. If purchase cost is low and critical cost is low to high and the part
is not readily available then order large quantities of stock for life
of mission.

3. If purchase cost is medium and criticality is low to medium or

purchase cost is high but criticality is low, then order in batches
at optimal order quantity and maintain a reasonable safety stock.

4. If purchase cost is high and criticality is medium to high, maintain

an absolute minimum+ 1 additional spare, and order frequently on
a one by one basis.

Material availability and location of the process from supply sources

will play a significant role in shaping the inventory policy. Fluctua-
108 MAINTENANCE, MODELING AND OPTIMIZATION

tions in failure rates and population of equipment size will need to be

mentioned to assure that proper inventory levels are established.
The ordering strategies discussed above cover both the safety stock
and initial stocking level. However, the costs and risks need to be bal-
anced in order to strive at an inventory stocking level for specific spare
parts. The approach can easily be integrated with the existing in-
ventory management system of any major organization, with minor
modifications. The reliability centered forecasting seamlessly integrated
with inventory control strategies is a useful approach for forecasting the
need of nonrepairable parts in a given planning horizon, and provide the
strategies of their effective maintenance.

ACKNOWLEDGEMENT: The authors acknowledge the support

provided by King Fahd University of Petroleum & Minerals.

References

[1] Al-Bahli, A. M. Spares Provisioning Based on Maximization of

Availability per Cost Ratio. Computers in Engineering, 24(1),
(1993) 81-90.
[2] AT&T, Reliability Manual. Basking Ridge, NJ;AT&T, 1983.
[3] Bartman, D., and Beckmann, M. J. Inventory Control; Model and
Methods. Springer Verlag, 1992.
[4] Berg, M., and Posner, M.J.M. Customer Delay in M/G/oo Repair
systems with Spares. Operations Research, 38(2) (1990) 334-348.
[5] Bloch, H. P. and Geitner, F. K. Practical Machinery Management
for Process Plants. 2, Gulf Publishing Company, Houston, 1983.
[6] Bolton, W. Production Planning and Control. Longman Scientific
& Technical, 85-96, 1994.
[7] Cooper,R.B. Introduction to Queuing Theory. Macmillan, New
York, 1972.
[8] Dhakar, T.S., Schmidt, C.P., and Miller, D.M. Base stock level
determination for high cost low demand critical repairable spares.
Computers Ops Res., 21 (1994) 411-420.
[9] Ebeling, C.E. Optimal Stock Levels and Service Channel Allo-
cations in a Multi-Item-Repairable Asset Inventory system. IIE
Transactions, 23(2) (1991) 115-120.
[10] Gnedenko, B. V., Belyayev, Yu. K. and Solovyev, A. D. Mathe-
matical Methods of Reliability. Academic Press, New York, 1969.
Reliability Based Spare Parts Forecasting and Procurement Strategies 109

[11] Graves, S.C. A Multi-Echelon Inventory Model for a Repairable

Item with one-for-one Replenishment. Mgt. Sc., 31(10) (1985)
1247-56.
[12] Greene, J. H. Production and Inventory Control Handbook.
McGraw-Hill book company, 1970.
[13] Gross, D. On the Ample Service Assumptions of Palm's Theorem
in Inventory Modeling. Mgt. Sc., 28(9). (1982) 1065-1079.
[14] Gross, D., Miller, D.R, and Soland, RM. On Some Common In-
terests among Reliability, Inventory and Queuing. IEEE Trans-
actions on Reliability, R-34(3) (1985) 204-208.
[15] Hall and Clark, A.J. ACIM: Availability Centered Inventory
Model. Proceedings of the Annual Reliability and Maintainabil-
ity Symposium, IEEE, (1987) 247-252.
[16] Kales, P. Reliability for Technology. Prentice-Hall, Inc., 1998.
[17] Langford, J. W. Logistics: Principles and Applications. McGraw-
Hill, Inc., N.Y. 1995.
[18] Lewis, E. E. Introduction to Reliability Engineering. John Wiley
& Sons, Inc., New York, 1996.
[19] Lipson, C. and Sheth, N. J. Statistical Design and Analysis of
Engineering Experiments. McGraw-Hill, New York, 1973.
[20] Mann Jr, L., Saxena, A., and Knapp, G.M. Statistical-based or
condition-based Preventive maintenance? J. of Quality in Main-
tenance Engineering, 1(1) (1995) 46-59.
[21] Monks, J. G. Operations Management. McGraw-Hill Book Com-
pany, 1985.
[22] Orsburn, D. K. Spares Management Handbook. McGraw-Hill Inc.,
1991.
[23] Petrovic, R, and Pavlovic, M. A New Performance Measure
for Multi-level Spare Parts Inventories. Inventory in Theory and
Practice, Elsevier, Editor Attilia Chikan, Elsevier, 1986.
[24] Petrovic, R Senborn, A., and Vujosevic, M. Hierarchical Spare
Parts Inventory Systems. Elsevier, 1986.
[25] Purser,F.E. and Farmer,G.F. Optimizing Spare Parts Inventories
Using RAM Techniques. Idaho National Engineering Laboratory
(EGf3G Idaho),ASME Paper 90-JPGC/Pwe-50,1990.
[26] Raza, M. K. Some quality and reliability aspects of aluminum
extrusion. M.S. Thesis, King Fahd University of Petroleum &
Minerals, Dhahran, 1999.
110 MAINTENANCE, MODELING AND OPTIMIZATION

[27] Sherbrooke, C.C. METRIC: a Multi-Echelon Technique for Re-

coverable Item Control. Operations Research, 16 (1968) 122-14l.
[28] Sherbrooke, C.C. Optimal Inventory Modeling of Systems. John
Wiley, New york, 1992.
[29] Sheikh, A. K., Callom, F. L. and Mustafa, S. G. Strategies in
Spare Parts Management using a Reliability Engineering Ap-
proach. Engineering Cost & Production Economics, 21 (1991)
51-7l.
[30] Sheikh, A. K. A Statistical Method of Calculating the Spare Parts
Requirement. Symposium on Maintenance, Planning and Opera-
tion, King Saud University, Riyadh, Saudi Arabia, March 17-19,
1990.
[31] Sipper, D. and Bulfin Jr. R. L. Production planning, control and
integration. McGraw Hill Companies, Inc., 1997.
[32] Younas, M. Reliability Models in Fatigue Life Prediction. M.S.
Thesis, King Fahd University of Petroleum & Minerals, Dhahran,
1984.