See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/45534747

Level of Repair Analysis based on Genetic Algorithm with Tabu Search

Article · June 2010

Source: DOAJ

CITATIONS READS

3 222

3 authors, including:

Taoufik Bouachera Mohammed Kishk

algerian petroleum institue iap University of the West of England, Bristol
4 PUBLICATIONS 31 CITATIONS 66 PUBLICATIONS 793 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Mohammed Kishk on 30 May 2014.

The user has requested enhancement of the downloaded file.

Proceedings of the World Congress on Engineering 2010 Vol III
WCE 2010, June 30 - July 2, 2010, London, U.K.

Level of Repair Analysis based on Genetic Algorithm

with Tabu Search
Taoufik BOUACHERA1, Mohammed KISHK1 and Laurie POWER2

Abstract - Genetic algorithms and their hybrid schemes have Since these support resources are very costly for systems
shown a great efficacy in solving large scale combinatorial containing thousands of components, there is always a trade-
problems in which solutions are highly time-consuming. The off between availability and maintenance costs. Ignoring these
level of repair analysis (LORA), mathematically formulised support issues in the early phase, however, may have costly
by an integer programming model (IP), is very difficult to consequences during the operation phase. This becomes in
optimize by means of traditional optimization techniques due increasingly evident with fierce market conditions. As a
to a large number of decision variables involved. In this paper, result, companies are compelled to move from fragmented
a hybridised Genetic Algorithm with Tabu Search is presented decision processes toward more integrated acquisition and
and its application to solve Level of repair analysis (LORA) operation decision in order to sustain their equipment use at
problem is investigated. The LORA, considered as an low cost and required availability. Consequently, some
important tool for strategic system maintenance decision industries become aware of the large potential for cost
making, seeks to determine the location in the repair network reductions by adopting whole life techniques in their
at which a failed component should be discarded or repaired. acquisition process. Level of repair analysis LORA is one of
The proposed algorithm is developed in order to determine the the prescribed techniques in the military and maritime
best repair decision combination. The efficacy of the industries to achieve a system design with the minimum
algorithm is investigated in the context of a case study. The whole life maintenance cost (MIL.STD.1390D).
maintenance costs of a structure of three-echelon repair and
multi-indenture is optimised under the condition that repair
decision should be taken for all system items. Typical results 1.2. Level Of Repair Analysis (LORA)
have shown that the algorithm can effectively handle a real
industrial sized case study with adequate optimisation The LORA approach was developed by military industry to
computational time. evaluate the Integrated Logistic Support (ILS) factors
contributing to the systems whole life cost. When a failure
Keywords: Level of repair analysis, maintenance occurs, failed components are removed and repaired or
optimisation, Genetic Algorithms, Tabu Search. replaced by new spare parts. In designing systems, Level of
repair analysts considers all aspects of the system design and
1. Introduction maintenance scenarios to achieve availability and cost
balanced systems. As a result, they provide essential support
1.1. Background requirements for the most effective maintenance strategy
Over the last decade considerable emphasis has been put on under predicted operational environments.
whole life costing (Tysseland, 2007; Kleyner & al., 2008;
Lindholm et al., 2004; Kishk et al., 2003). Traditional system The basic of the LORA process is the following. Level of
acquisition has taken place on the purchase costs without repair analysts have to decide for a given design which
explicit indication of operation costs. However, the decrease components to repair, which components to discard, where in
of company operational budgets has unveiled the need for the repair network to do this and finally where in the repair
techniques to forecast and to optimise ongoing cash flows network to install the required maintenance resources. Thus, a
over the system whole life. As result, a prerequisite for number of reparation locations in which systems, subsystems
effective acquisition decision is the operational parameters and components have to be repaired or discarded is set up to
such as: system availability, system reliability and the cost of satisfy maintenance requirements at minimum cost.
the required support resources. Besides, it was found that
operational readiness of expensive complex structures such as
petroleum apparatus, aircrafts, ships and military equipments
is very sensitive to the availability of spare components,
maintenance resources and manpower.
1
The Scott Sutherland School, The Robert Gordon University, Garthdee
Road, Aberdeen AB10 7QB, UK.
2
School of Engineering, The Robert Gordon University, Schoolhill, Aberdeen
AB10 1FR, UK.
Figure 1 : A multi-echelon repair network

ISBN: 978-988-18210-8-9 WCE 2010

ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
Proceedings of the World Congress on Engineering 2010 Vol III
WCE 2010, June 30 - July 2, 2010, London, U.K.

In the literature, various models of LORA have been discarding it. Level of Repair Analysis (LORA) is an
discussed for a three echelon repair network (figure 2) and approach which examines the cost balance between repairing
multi indenture system (figure 1). These models involve a the component and discarding it. The framework of this
large number of decision variables which makes LORA approach is an iterative process ensuring the optimum
problem very difficult to optimize by means of traditional maintenance planning. However, the LORA problem as
optimization techniques. For instance, the number of all combinatorial optimization is not the most widely studied in
possible combinations (part, repair and discard decision) for a the literature (Basten & al., 2006). A limited works were
system consisting of 32 parts spread between different devoted to solve the LORA problem. Barros et al. (2001),
indentures is 6.28 × 1010 (Kumar & al., 2006). Hence, Saranga et al. (2006), Gutin et al. (2005) and Basten et al.
techniques like integer programming and branch and bound (2009) modelled LORA as Integer Programming model in
method become difficult to use. Consequently, we focus on which all repair locations at the same echelon were
the opportunity of using Genetic Algorithms which are the aggregated. Besides, they all resolved their model under
most suitable to problems involving combinatorial infinite capacity of resources. Brick et al. (2009) model
optimisations. LORA without aggregating data per echelon level for only 1
echelon and 2 indenture levels.

Barros & al. (2001) presented a mathematical framework

as an Integer Programming (IP) model resolved by branch and
bound algorithms. In this model the objective function has
two elements: a fixed costFC of setting up maintenance
facilities (test equipment, labour manpower and technical
data); and a variable cost CV of ordering and holding spare
parts. The goal of this IP problem is to find a subset of repair
decisions that minimises the total maintenance costs while
satisfying parts relationship and maintenance resource
Figure 2 : A multi-indenture system
constraints. They have assumed that any installed
maintenance capacity (fixed cost) performs reparation to all
components belonging to the same indenture. Brik et al.
1.3. Objective & methodology
(2009) work treated the applicability the location of facilities
and installation of capacitated resources to LORA problems.
Based on the discussions above, this paper addresses the issue They have proposed a mixed-integer problem MIP model for
of minimising maintenance costs based on level of repair the discrete location of facilities and installation. Gutin et al.
analysis. This optimisation is modelled as integer (2005) formulate the LORA problem as an optimization
programming (IP) model in which the space solution is homomorphism problem on bipartite graphs and they have
proportional to the number of echelons and the system proved that the LORA problem a NP-hard problem. Saranga
structure size. In addition, not all maintenance features are & al. (2006) adopt the same Barros Integer Programming
studied, but the study is restricted only to model the effect of model but with different fixed cost allocation. They
repair capacity on whole life maintenance cost. Optimal considered that any component bears a specific fixed cost
LORA approaches often make use of heuristic methods to whereas in Barros model all components at the same
optimize the IP model. In this sense, the main aim of this indenture share the same fixed costs. Furthermore, Sarraga et
study is to show that Genetic Algorithms and Tabu Search can al. have solved LORA problem by using the genetic algorithm
yield to optimal solution at an acceptable computing time software evolver. Basten & al., (2009) propose an Integer
with respect to solution space. The obtained algorithms can be Programming model that generalizes the existing models
seen as building blocks for more general maintenance (Barros model and Saranga model) by allowing a predefined
optimisation model. set of components to share the same fixed costs. In addition,
they modelled the LORA problem as a minimum cost flow
The remainder of the paper is organized as follows. Section 2 problem with side constraints.
presents related works and a mathematical formulation of the
LORA problem. Section 3 describes Genetic Algorithm, Tabu In this paper, we model the LORA problem as described in
Search and their combination GATS algorithm used to solve Barros, Sarranga and Basten models. We obtain the optimum
the level of repair analysis (LORA) problem. Section 4 repair decisions by minimising the maintenance cost given by
presents a numerical application of the proposed algorithm. equation (1) under the constraints (2, 3 and 4). The added
Section 5 concludes the paper and offers some suggested value of this paper to the literature is that we used a hybrid
directions for future research. Genetic and Tabu Search algorithms which are very suitable
to NP-hard problems and combinatorial optimisation.

2. Related works The following notations are used in our model.

m = the number of the echelons in the reparation network.
The most economical maintenance strategy for any n = the total number of components for the system under
component of system is to decide it is worth repairing or consideration.

ISBN: 978-988-18210-8-9 WCE 2010

ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
Proceedings of the World Congress on Engineering 2010 Vol III
WCE 2010, June 30 - July 2, 2010, London, U.K.

Component i is the parent of the component j or component algorithms to enhance their capabilities in solving
j is the child of the component i combinatorial optimisation (Zdanski & al., 2002 and
r = repair options: repair, discard or move. Hagemana & al., 2003). For instance, a GA speed is low for
i = Total number of maintenance tasks required in the the huge size population and TS relies strongly on the initial
whole life time of component i. solution. Consequently, GA and TS combination named
CFr,e,i = fixed cost related to repair option ‘r ’ at echelon e, GATS may overcome these limitations and maintain their
for component i. advantages.
VC r,e,i = variable cost related to repair option ‘r’ at echelon
e, for component i.
3.1. Genetic algorithms
Let X, the repair decision, is 1 if repair decision has been
chosen at the echelon e of the selected component i and 0 Genetic algorithms are stochastic search techniques based on
otherwise. the theory of evolution for finding the global optimum
solution. The genetic algorithm developed by Holland to
1 if repair option r at echelon e is selected for part i optimise a function F(x), where x is a vector representing
X r,e,i  
0, otherwise individual solutions (Gen & al. 2000). First of all, Genetic
algorithms generate not only a single solution but a group of
Using the notation mentioned above, the total maintenance solutions, called a population. This population changes over
cost is: time, but it always keeps its initial size. The population
members are called strings or chromosomes from which a
 VC it  FC r,e,i X r,e,i
N 3 m

r,e,i
subset called parents is selected according to the best values
i 1 r 1 e 1 of F(x). A fitness value in Genetic algorithms is a measure of
goodness of a solution to the objective function, i.e., the
(1) fitness of an individual is directly related to its objective
Subject to function value. At any iteration, a fitness value is calculated
X r, 1, i  1 for all parts for each of the current individuals. The selection rule, called a
survivability test, exclude from the population the strings
which have the worst finesses. Second, new solutions called
(2) children (or offspring) are produced by genetic operators:
3 crossover and mutation. Together parents and new children
X move, e, i  r 1
X r, e  1, i  1 are grouped in a new population which will pass again
through survival test. Thus, the population as a whole moves
iteratively towards better solutions ideally to the global
(3) optimum.
X r, e, j  X r, e, i

 e and (i is parent of j) where r  discard or move Chromosome representation

(4)
The first step in implementing a genetic algorithm for a
particular problem is to adopt a suitable chromosome
The objective function given in Eq. (1) sums the fixed and
representation. The representation scheme developed in this
variable costs of performing repair and discard actions. The
paper was a (n x d) binary matrix, where n is the number of all
constraint given in Eq. (2) ensures that one repair option is
parts under consideration and d is the number of all the repair
chosen at the echelon one. If a move decision is taken at the
decisions throughout the repair network. A value of 1 in this
echelon e, only one repair decision should be taken at the
representation implies that a repair, discard or move decision
echelon e+1 (constraint given in Eq. 3). Otherwise, no repair
has been attributed to the component i and the echelon j. The
option is chosen at the echelon (e+1).The equality constraint
binary representation of any chromosome or solution is
given in Eq. (4) requires that all the enclosed lower indentures
visualised in Figure 3.
of any subsystem have the same decisions of the subsystem
itself with respect to the replace and move options at different
Furthermore, any technical system may be considered as
echelons. The last constraint requires that there are only two
collection of assemblies which are in turn considered as a
repair decisions (repair or discard) at the last highest echelon.
collection of a set of subassemblies. This perspective on
Figure (3) represents a sample of possible solution generated
technical system is illustrated in figure 2. The number of
randomly by taking into consideration all the above
levels, also referred as indenture levels, in the material
constraints.
breakdown structure of technical system is limited to the
deeper detailed information needed for repair tasks and spare-
3. Hybrid Genetic & Tabu Search algorithms
part provision.
Either the genetic algorithm GA or the tabu search TS are
Fro a modelling perspective, the system breakdown structure
suitable tools for solving such problems. In the literature,
is represented by a matrix, referred in the literature by
however, several researchers have tried to combine these two

ISBN: 978-988-18210-8-9 WCE 2010

ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
Proceedings of the World Congress on Engineering 2010 Vol III
WCE 2010, June 30 - July 2, 2010, London, U.K.

commonality matrix (figure 4), where the column represents generated randomly. In addition, we select one of the best
parent items and in the row are child items. We start by solutions and we generate a repair decision for a component
assorting parts from the first indenture until the last but one selected at random. Again, we adjust the new changes
indenture in the column as parent items. Then, we insert parts according to the constraint Eq. (4).
from the second indenture to the last one in the commonality
matrix row. As shown, child parts 5, 6 and 7 belong to parent In our GATS algorithm, these two operators are applied for
part 3 or parent part 3 is constituted of child parts 5, 6 and 7. the individual generated by Genetic Algorithm and improved
According to this representation, whenever the parent part 3 is by tabu search.
under discard or move decision, the child parts 5, 6 and 7 will
have the same decision (constraint Eq. 4).

3.2. Tabu Search

Tabu Search, concept based on the use of memory, tries to

keep track of solution already visited. By leading the
optimisation to new areas, TS is able to attain the global
optimum instead of local minima. The framework of TS
consists of generating some neighbouring solutions from an
initial solution (Eswaramurthy & al., 2009). These solutions
are evaluated by means of objective function and sorted. The
Figure 3: Sample of repair decision tabu list is updated by the best solution according to its
fitness. Afterwards, a new solution is identified and additional
neighbouring searches are generated from it. When the best
solution remains unchanged after a number of iterations, the
optimum is achieved and the best solution will be returned.

Figure 4: Matrix representation for system structure

Genetic Algorithm operators

The GATS algorithm uses fitness proportional selection with

roulette wheel sampling for crossover operator. At each
generation Elitism is applied in this study by replacing the
worst solution by the best one with respect to total cost given
in Eq. (1). After a pair of parents is selected, the crossover
operator produces two new children or off springs. The
crossover operator is applied on these two parent
chromosomes by interchanging the information extracted
from them. Since each parent’s genetic code has the same
structure, we apply one-point the crossover by considering the
same crossover point selected at random. The children are
generated by combining the left and right parts (figure 2);
which is followed by adjusting the offspring repair decisions
with respect to the constraint Eq. (4).

On the other hand, mutation is the other important element in

genetic algorithms that creates randomly new children. This
operator serves as a strategy to prevent solutions from being Figure 5: An example use of the crossover operator
trapped in local optima. In this work, the mutation operator
works by selecting randomly one chromosome outside the
best solution list and replacing it by a new chromosome also

ISBN: 978-988-18210-8-9 WCE 2010

ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
Proceedings of the World Congress on Engineering 2010 Vol III
WCE 2010, June 30 - July 2, 2010, London, U.K.

The procedure of TS consists of the following steps which are 3. Repeat step 2 until there is no improvement of the best
depicted in the figure (6). First, a number of neighbourhood fitness value.
solutions that can be produced from an initial solution are 4. Replace the solution by its best neighbourhood.
examined. Then, a solution with the best fitness value and it is 5. Choose two solutions to produce new chromosomes
not in the tabu list is selected from the explored using genetic operators: parent selection and crossover.
neighbourhood. This way, tabu search tries to assure that the These new solution are accepted when they verify the
method does not re-examine a solution previously generated. constraint equations 2, 3 and 4.
Finally, TS procedure iterates the previous step until no more 6. Create a new chromosomes using genetic operator:
neighbours are present (all are tabu), or when during a mutation.
predetermined number of iterations no improvements are 7. Update a tabu list of the best chromosomes.
found. 8. Repeat step 1 until there is no improvement of the best
chromosome.

The proposed algorithm has been implemented into a

computer routine using the MATLAB® programming
environment (The MathWorks, 2008).

4. Computational experience

In this section, we present the results of numerical

experiments to test the effectiveness of our GATS algorithm.
For comparison sake, we conduct first the same case study
that was carried out in Saranga’s work (Saranga & al., 2006).
Figure 6: General flowchart of the GATS algorithm

3.3. Genetic & Tabu Search Algorithm Table 1: Case study data used in Saranga’s work, maintenance
costs
This approach, widely used in the literature, combines the
advantages and mitigates the disadvantages of the two
algorithms. Tabu search relies only on one solution and miss
information of a larger set of solutions, however, Genetic
Algorithms lead to lower solution quality with increasing
problem size (Zdanski & al., 2002). In this study the GATS
algorithm starts by generating N initial possible solutions
(figure 6). A tabu search, as an iterative process, is then used
for upgrading theses solutions through neighbouring
exploration. Afterwards, the flow returns to the Genetic
Algorithm which is again an iterative process. By means of
the genetic operators new off springs are produced. Then, a
tabu list of the best solutions is updated by the new off springs
according to the fitness value. The stopping criteria for the
GATS algorithm are a predefined number of consecutive
iterations attaining the same best solution is reached.

The mean steps of the algorithm are shown in figure 6 and Table 2 : Case study data used in Saranga’s work, material
described as follows: structure
1. Generate randomly a set of solutions (20 solutions)
verifying the equations 2, 3 and 4.
2. Refine each solution by the neighbourhood routine with
respect to fitness value. A neighbourhood solution is
obtained only by modifying the value of one element
from the solution under consideration to 1 or 0. Besides,
the neighbourhood solutions are not accepted until they
verify the constraint equations 2, 3 and 4. Then, a tabu
list is updated containing all the fitness values of the In this experiment, the settings are chosen as described in
solutions that have been explored. After, a new (Saranga & al., 2006) on two echelon repair network for an
neighbourhood is explored only when its fitness value aircraft engine with three-indenture structure. Different costs
does not exist to the tabu list. (CF: fixed costs and CV: variable costs) for various repair

ISBN: 978-988-18210-8-9 WCE 2010

ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
Proceedings of the World Congress on Engineering 2010 Vol III
WCE 2010, June 30 - July 2, 2010, London, U.K.

options at different echelons for all items are listed in Table A comparison between two and three echelon network
(1). Besides, the commonality matrix shows the relationship computational time that takes GATS algorithm to come out
between first indenture parts (from 1 to 10) and the second with the optimal solution is shown in (figure 7). The
indenture parts (from 11 to 32). The optimal or near optimal computational time increases exponentially with system
solution obtained by Saranga’ work and our GATS algorithm structure size and the bigger the number of echelon is the
are similar, only part 5 got different repair decision (table 3). higher the computing time is. Thus, researchers consider
The total maintenance costs incurred are respectively three echelon repair network is enough in practice to handle
4255.274 and 4216.274. In addition, the maximum maintenance activities and to be modelled by acceptable
maintenance cost achieve by the simulation is 42 759 computational time.
representing more than ten times the optimum solution. This
witnesses the benefit of adopting LORA in maintenance
planning and support provision.

As a second important issue related to the optimisation

problem is the computational time. The algorithms GATS is
written in the MATLAB language and implemented on a
Pentium 4 CPU 2.60 GHZ with 512 Mo RAM. The
computing time required to solve the LORA problem varies
with system structure (total number of items) and the repair
network. Figure 7 gives the computing time taken to solve the
problems for the data sets created randomly for 3 echelon
network. For problem that has been discussed above, it took Figure 7 : GATS computational time
an average time of 21 seconds to solve the problems. As was
previously mentioned, the solution representation is a (n x d)
binary matrix, where n is the number of all parts under
consideration and d the number of all repair decision 5. Conclusion and future works
throughout the repair network. The solution has for a system
with n parts with m echelons and ri repair options at echelon i, A typical level of repair analysis includes multiple repair
the number of possible solutions is equal to 2n*s. facilities and thousands of system items, is formalised by
Integer Programming (IP) model. Traditional optimisation
m techniques can not be effectively applied to solve LORA
Where: s  r
i 1
i , ri is the number of repair options models for real-world applications in which systems may
enclose millions of parts. In this paper, a hybrid algorithm of
at echelon i. Genetic Algorithm and Tabu Search (GATS) has been
For a case study with 40 parts, the size of the solution space developed and implemented into a computational algorithm in
will be for 3 echelons as high as 2,14 × 1096. MATALAB code to solve this mathematical formulation. The
algorithm adopts a matrix representation for the system
Table 3 : Best LORA solution for Saranga’s case study breakdown structure to handle the constraint linking parent
items and children items. The efficacy of the algorithm has
been validated in the context of an example application. The
repair decision of all system items has been optimised for a
structure of three-echelon repair network and multi-indenture
system. The results have shown that quite large LORA
optimisation can be obtained in realistic times, demonstrating
that the algorithm is practical.

There are some practical issues that need to be addressed,

however. This LORA problem optimises maintenance costs
based only on repair facilities. This should be extended to
include other maintenance costs such as spare part provision.
Further research in this area will include studying the impact
of both spare part provision and repair facilities on LORA
problems. Besides, spare part optimisation under finite repair
capacity is being integrated into the development of the
algorithm and will be reported in a future paper.

ISBN: 978-988-18210-8-9 WCE 2010

ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
Proceedings of the World Congress on Engineering 2010 Vol III
WCE 2010, June 30 - July 2, 2010, London, U.K.

References of aircraft maintenance/support infrastructure using

genetic algorithms - level of repair analysis”. Annals of
1. Alfredsson, P., 1997, “Optimization of multi-echelon
Operations Research.
repairable item inventory systems with simultaneous
location of repair facilities”. European Journal of 17. Tysseland, B. E., 2007, “Life cycle cost based
Operational Research, 99:584 procurement decisions A case study of Norwegian
Defence Procurement projects”, International Journal of
2. Barros, L. and Riley, M., 2001, “A combinatorial
Project Management.
approach to level of repair analysis”, European Journal of
Operational Research 18. Zdanski, M. And Pozivily, J., 2002, “Combination
genetic/tabu search algorithm for hybrid flowshops
3. Basten, R. J. I., Schutten, J. M. J., and Van der Heijden,
optimization”, Proceedings of Algoritmy, Conference on
M.C., 2009, ”An efficient model formulation for level of
Scientific Computing.
repair analysis”. Annals of Operations Research.
19. The MathWorks (2008) Using MATLAB®. The
4. Brick, E.D. and Uchoa, E., 2009, “A facility location and
MathWorks, Inc., version 7.6.
installation of resources model for level of repair
analysis” European Journal of Operational Research,
192(2):479-486.
5. Eswaramurthy, V. P. and Tamilarasi, A., 2009,
“Hybridization of Ant Colony Optimization Strategies in
Tabu Search for Solving Job Shop Scheduling Problems”,
International Journal of Information and Management
Sciences
6. Gen, M. and Cheng., R.,2000, “Genetic Algorithms and
Engineering Optimization”. John Wiley & Sons, Inc.
7. Glover, F. and Laguna, M., 1997, “Tabu search”, Kluwer
Academic Publishers, 1997
8. Goldberg, DE., 1989, “Genetic algorithms in search,
optimization and machine learning”. Reading, MA:
Addison-Wesley.
9. Gutin, G., Rafiey, A., Yeo, A., & Tso, M., 2005, “Level
of repair analysis and minimum cost homomorphisms of
graphs”. Discrete Applied Mathematics.
10. Hagemana, J.A.; Wehrens R.; Van Sprang H.A. and
Buydens L.M.C, 2003, “Hybrid genetic algorithm–tabu
search approach for optimising multilayer optical
coatings, Analytica Chimica Acta 490
11. Jones, JV., 2006, “Integrated Logistics Support
Handbook”, McGraw-Hill; NewYork, 3rd edition.
12. Kishk, M.; Al-Hajj, A.; Pollock, R.; Aouad, G.; Bakis,
N.; and Sun, M., 2003, “Whole life costing in
construction A state of the art review”, Published by
RICS Foundation, London, UK.
13. Kleynera, A. and Sandborn, P., 2008, “Minimizing life
cycle cost by managing product reliability via validation
plan and warranty return cost”, International Journal of
Production Economics
14. Lindholm A., and Suomala P., 2005: ‘Present and Future
of Life Cycle Costing: Reflections from Finnish
Companies’, Canadian Journal of Civil Engineering 32:
250–259.
15. MIL.STD.1390D, 1988, “US Military Standard: Level of
repair analysis”. US department of defence.
16. Saranga, H. and Dinesh Kumar, U., 2006, “Optimization

ISBN: 978-988-18210-8-9 WCE 2010

ISSN: 2078-0958 (Print); ISSN: 2078-0966 (Online)
View publication stats