Professional Documents
Culture Documents
https://doi.org/10.1007/s12351-019-00540-x
ORIGINAL PAPER
Abstract
This paper examines a discrete material two echelon supply chain system. Multi-
ple reliable non identical merging suppliers send products to an intermediate stor-
age area which can in turn feed a distribution centre with parallel identical reliable
distribution channels. It is assumed that each merging supplier may have parallel
identical reliable supply channels. The service rate of each supplier and each iden-
tical channel at the distribution centre is assumed to be exponentially distributed.
The examined model is analyzed as a continuous time Markov process with discrete
states. An algorithm that can create the system’s transition probabilities matrix for
any value of its parameters is presented and various performance measures are cal-
culated. The comparison of the proposed method with simulation showed that the
proposed algorithm provides very accurate estimations of the system’s performance
measures. Additionally the optimal values of the system’s parameters to optimize its
various performance measures are also explored thoroughly.
* Alexandros Diamantidis
adiama@econ.auth.gr
Michael Vidalis
mvid@ba.aegean.gr
Stelios Koukoumialos
skoukoum@teilar.gr
George Blanas
blanas@teilar.gr
1
Department of Business Administration, University of Aegean, Chios, Greece
2
Department of Business Administration, T.E.I. of Thessaly, Larissa, Greece
3
Department of Economics, Aristotle University of Thessaloniki, Thessaloniki, Greece
13
Vol.:(0123456789)
704 M. Vidalis et al.
A supply chain (SC) usually has several echelons that can be connected to each
other. The performance of an echelon can affect directly the performance of the
rest echelons of the SC. The main target of the supply chain management (SCM)
is to control the flows between its stages in order to maximize the total profit-
ability of the SC. Usually the parameters of the SC that affect its performance and
behavior are the number of suppliers, the distribution centers capacities and the
capacities of the intermediate storage areas. Another important factor is the exist-
ence or not of stochastic processes that describe the service rates of the suppliers
and the distribution centers.
The randomness of the processes involved makes necessary the use of a sto-
chastic evaluative model that can estimate various performance measures. Evalu-
ative models make use of input data such as the system parameters and provide
as an output the performance evaluation of the system. The evaluative models
can be combined with generative models to create a tool for the optimal design
of production systems or supply chains. The goal of this study is the creation of
an evaluative model for the estimation of the considered system’s performance
measures.
A push based supply chain with suppliers, intermediate buffers/storage areas
and distribution centres, can be viewed as a push flow line where the suppliers of
the supply chain correspond to the workstations of the flow line and the distribu-
tion centres of the supply chain, correspond to workstations with parallel serv-
ers of the flow line. As a result of this correspondence, the analysis of a supply
chain can be done using the same techniques and literature with those used for the
analysis of manufacturing systems. Therefore the considered system can be con-
sidered as a two-stage queuing network with a finite storage area that can be also
used for production, telecommunications, and also for inventory management if
the assumptions can be justified depending on the application. Moreover merging
suppliers can be used to model and analyze assembly supply chains where com-
plementary parts from different suppliers can be assembled together into a new
product and then distributed through a distribution centre to the final customers.
In contradiction to the classical serial systems, merge systems haven’t been
studied thoroughly in the literature,. Moreover, the analysis of push merge sys-
tems where each merging workstation may have parallel servers draws a little
attention in the literature.
Regarding the analysis of systems with merge operations, Tan (2001) studied a
three stations push model with a unique storage area, merge operations and con-
tinuous flow of material. The times to failure and to repair were exponentially
distributed. The methodology used for the analysis of the considered model was
renewal theory in conjunction with level crossing analysis.
Helber and Mehrtens (2003) examined a three stations one buffer merge sys-
tem for the case of continuous flow of material. They assumed machine specific
processing times, while the times to failure and repair were assumed to be expo-
nentially distributed.
13
Analysis of a two echelon supply chain with merging suppliers,… 705
Diamantidis et al. (2004), analyzed using a discrete state, discrete time Markov
chain a three machine one buffer merge system with discrete flow of material. They
assumed that the processing times were fixed and equal for all machines. Geometri-
cally distributed failure and repair rates of the machines were also assumed. A simi-
lar model with non identical merging machines which have exponential failure and
repair rates was examined by Quazene et al. (2013). Two transformation techniques
were used and compared. The first technique replaced the parallel machines on the
first stage by an equivalent machine, while the second one used overlapping decom-
position approach which decomposed the original line into two serial lines.
Liu and Li (2009, 2010) analyzed systems with split and merge operations operat-
ing with Bernoulli and exponential machine reliability assumptions using various
processing policies. They derived the recursive procedures for the performance anal-
ysis and they proved the convergence of the procedures along with the uniqueness of
the solutions.
Li et al. (2009), summarized recent studies in the area and they discussed approx-
imated methods to analyze complicated systems, with various topologies including
assembly/disassembly, split and merge and closed-loop systems. Moreover they pro-
posed future research topics from the automotive production lines.
Vidalis et al. (2012), considered a merge supply chain with two non identical
reliable suppliers a distribution center and a shared buffer. The supply rate of each
supplier and the service rate of the distribution center were assumed to follow the
Erlang distribution. They proposed an analytical model based on a discrete state,
continuous time Markov chain in order to evaluate various performance measures of
the examined system.
Geranios et al. (2016), examined a merge network with up to three unreliable sup-
pliers a distribution center and a shared buffer. The failure and repair times of the
suppliers were assumed to be Coxian-2 distributed. Two algorithms were proposed
that were used as a decision making tool to determine the optimal system’s configu-
ration in order to optimize its performance measures.
Two echelon discrete material supply chains with exponential processing rates at
all nodes were examined by Vidalis et al. (2014, 2015). They used continuous time-
discrete state Markov processes for the performance evaluation of the considered
systems. They proposed the appropriate algorithms to generate the transition matrix
for any values of the system’s parameters in order to estimate the performance meas-
ures of the examined systems.
Diamantidis et al. (2016), analyzed a push pull merge supply chain with two buff-
ers, distribution center and Poisson distributed external demand using a continuous
time, discrete state Markov chain. An algorithm for the performance evaluation of
the considered system was also proposed.
Chen et al. (2014), examined the impact of supply chian power structure on prof-
itability in an assembly supply chain with two suppliers and one assembler. They
investigated two power regimes regarding the relationship between the suppliers and
the assembler.
Modrak and Marton (2013) presented a theoretical framework for modeling of
assembly supply chains and proposed three possible indicators to measure a struc-
tural complexity of assembly supply networks. In this direction, Hamta et al. (2018),
13
706 M. Vidalis et al.
13
Analysis of a two echelon supply chain with merging suppliers,… 707
Baumann and Sandmann (2017), examined two station tandem queues with mul-
tiple identical reliable servers at each station. The service times were assumed to
follow phase type distribution. They provided an exact computational analysis of
various steady state performance measures by modelling the tandem queue as a level
dependent quasi birth and death process.
Diamantidis et al. (2007), analyzed using a discrete state, continuous time
Markov process, small subsystems with two stations and one buffer that were used
later as decomposition blocks for the analysis of large systems.
The performance evaluation of systems with reliable multi server stages and
finite intermediate buffers was presented by Shin and Moon (2014). They examined
two-stage tandem queues with two buffers using the level dependent quasi-birth-
and-death process.
Wu et al. (2014), examined multi server queuing systems, with unreliable serv-
ers. They used quasi birth and death process to model the system. They calculated
approximately the rate matrix and the steady-state stationary distributions were
obtained by a matrix-analytic approach.
Shin and Moon (2017), analyzed first small systems with three multi server work-
stations and two buffers using a continuous time Markov process. Later these sys-
tems were used a decomposition blocks for the analysis of large systems.
The methodology presented in this paper differs significantly from the existing
techniques in the literature because it does not use replacement techniques to replace
the parallel machines at each merging workstation with an equivalent single server
merging workstation. Moreover the considered system assumes parallel servers at
each work station which makes the system more complicated and to our best knowl-
edge such a system has not been examined in the literature so far.
This paper has the following structure: the assumptions of the model and the
solution procedure are given in Sect. 2. Section 3 describes the form of the transition
probabilities matrices and Sect. 4 presents all the necessary steps of the proposed
algorithm. Section 5 presents the mathematical formulas of the examined perfor-
mance measures, while Sect. 6 examines the impact of each system parameter on its
performance measures. Finally Sect. 7 concludes the paper and presents some ideas
for further research.
This study examines a push supply chain network managing discrete materials con-
sisting of non identical merging suppliers each one contains reliable parallel supply
channels, an intermediate storage area and a distribution centre (DC) with parallel
identical reliable channels.
The theoretical and practical contribution of this research is that it provides an
analytical tool that can be used for the performance evaluation and optimization of
merge-assembly multi echelon supply chains. In real life, merge-assembly supply
chains are extensively used in manufacturing industries, particularly in the automo-
bile, computer and electronic industries. One of the major challenges at the early
configuration design stage is to make a decision about the most suitable topology
13
708 M. Vidalis et al.
(i.e., number of suppliers, replenishment rate, storage area capacity etc.). The pro-
posed analysis can help the practitioners to make optimal decisions in this direction.
As depicted in Fig. 1, a number of M, (M ≥ 1) non-identical suppliers, send prod-
ucts (of one type) to a storage area with capacity Α that is located immediately
downstream. Each one of the j, j = 1,…,M, non identical merging suppliers may have
a number of identical parallel supply channels Pj (usually called capacity of supplier
j). The total processing rate of the jth supplier, j = 1,…,M is exponentially distrib-
uted and is denoted by μj, j = 1,…,M, respectively.
Let 𝜇i j , i = 1, … , Pj , j = 1, … , M , denotes the exponentially distributed process-
ing rate of the ith parallel supply channel at the jth supplier. It is straightforward that
∑P1 ∑P2 ∑Pk
𝜇 = 𝜇1 , i=1
i=1 i1
𝜇i2 = 𝜇2 , … , i=1 𝜇iM = 𝜇M . Additionally because all parallel
supply channels Pj at the jth supplier j, j = 1,…,M are identical it holds that
𝜇j
𝜇1 j = 𝜇2 j = ⋯ = 𝜇Pj j = Pj
, j = 1, … , M.
The last equation shows that all the parallel supply channels Pj of the correspond-
ing jth supplier, have the same exponential rate which is equal to Pj , j = 1, … , M .
𝜇
j
This equality leads to fewer numbers of states of the Markov chain. In that way the
system analysis becomes more manageable.
Because each supplier sends products to the storage area independently of the
other suppliers, the exponential distribution is used to model the processing rate of
each supplier. Moreover the use of the exponential distribution simplifies the exam-
ined complex model and is used often in the literature.
A distribution centre located immediately downstream of the shared storage area,
executes an additional processing on the items stored in the storage area. After this
processing the finished goods move outside the network.
It is also assumed that the suppliers are saturated i.e. there is always material
before the suppliers, and the DC is never blocked i.e. there is always free space area
after the DC in order to place the finished goods.
The distribution centre (DC) has a number of C (C ≥ 1) parallel identical reliable
channels where each one of the C identical parallel channels operates with an expo-
nentially distributed processing rate μ.
Suppliers DC
μ1
P1 μ
Storage area
μ
P2 C
μ2
PM μM
Fig. 1 A supply network with M suppliers, Pj parallel supply channels in each supplier, j = 1,…,M, a
shared storage area and a DC with C identical channels
13
Analysis of a two echelon supply chain with merging suppliers,… 709
When one supplier finishes processing a part and the storage area is full, then the
supplier is assumed to be blocked (blocking after service, BAS). Blocking stops if a
part is removed from the storage area because one channel at the DC finishes its ser-
vice. If more than one supplier is blocked, then the supplier with the smallest index
j, j = 1,…,M has the priority over the others to become unblocked first.
The considered system is described by parameters M, Pj, μj (j = 1,…,M), A, μ and
C, where M: number of merging suppliers; Pj: number of parallel supply channels at
the jth supplier; A: capacity of shared storage area (# of spaces); C: capacity of dis-
tribution centre (# of channels); μj: the total processing rate of the jth supplier; μ: the
processing rate of each channel at the DC.
The system depicted in Fig. 1, has significant practical value, since it can be used
to model real supply chain systems with merge operation that can be met in modern
supply chain networks. Moreover this system can be also used as a subsystem of
larger supply chains whenever merge operations occur at a specific point of the sup-
ply chain.
The proposed algorithm can handle theoretically systems (depending on hard-
ware capabilities) with any number of merging suppliers M, any number of parallel
supply channels Pj at each supplier j, any number of capacity of the storage area A
and any number of parallel channels at the DC.
The main goal of this study is the performance analysis of the system depicted in
Fig. 1. Additionally, we examined the influence of all parameters on its performance
measures. This has as a result the determination of the best combination for the sys-
tem parameters that optimize its performance measures. The optimization and the
performance evaluation of a such complicated system like the one depicted in Fig. 1
is the contribution of this paper.
2.1 Solution procedure
The examined model is described by a Markov process with continuous time and
discrete states. The form of the transition probabilities matrix which describes the
transitions among the states of the system, is affected by all the system parameters.
The system depicted in Fig. 1, is modeled as a birth–death stochastic process.
Births model the entry to the storage area and the DC and deaths model the exit from
the DC. As a result of this, the matrix of the transition probabilities is composed of
three different sets of matrix-blocks: (a) a set of nearly replica matrix-blocks in the
diagonal, denoted by ΜDi, (b) a set of nearly replica matrix-blocks above the diago-
nal, denoted by ADi, and (c) a set nearly replica of matrix-blocks under the diagonal,
denoted by UDi. So, the structure of the matrix of the transition probabilities has
a complete repetitive pattern. This repetition gives the opportunity to determine a
recursive generation for the matrix of the transition probabilities given any system
parameters. As a result, we can find the full set of the linear flow balance equa-
tions, then all the stationary probabilities, and finally all the system’s performance
measures.
A brief description of the steps for the performance evaluation of the examined
system is given below:
13
710 M. Vidalis et al.
The form of the matrix of the transition probabilities is affected by all four sys-
tem parameters. In order to find the steps that reveal the form of the final matrix
of the transition probabilities, initially the form of the matrix of the transition
probabilities when each one parameter changes at a time is explored. Finally, the
effect of all system parameters is combined by composing all the transition matri-
ces simultaneously in order to create the complex final matrix of the transition
probabilities.
Before we proceed to the description of the transition matrix it is useful to
make a reference to the simplest system that comes from the easiest combina-
tion of the four parameters, which is also the “first stone” for the entire matrix of
the transition probabilities. This system consists of one supplier (M = 1) with the
smallest number of supply channel (P1 = 1), no storage area capacity (A = 0) and a
single channel in the DC (C = 1). Generally, the number of states of a system with
M suppliers, P1, P2,…, PM, parallel supply channels at each supplier, A spaces
in storage area and C identical channels at the distribution centre is denoted by
SM(P1 ,P2 ,…,PM ),A,C . The number of states for the system (M = 1, P1 = 1, A = 0, C = 1),
i.e. the S1(1),0,1 is 3. Each state denotes the number of material units that being
served or waiting for service to the corresponding echelon/stage of the current
network.
In the above case there are two states for the supplier, i.e. the supplier is func-
tional (1), or blocked (b), and two states for the DC, i.e., idle (0) or functional (1).
When the supplier is in state block due to functional state of the DC, the material
unit belongs to the DC and not to the supplier. This state is denoted as b2.
So, the states of the system (1(1),0,1) are:
13
Analysis of a two echelon supply chain with merging suppliers,… 711
The matrix of the transition probabilities for the above system (1(1),0,1) is called
the fundamental matrix and has the following form (as shown in Table 1).
From the above description the matrix of the transition probabilities (MΤP) of the
system (1(1),0,1) can be presented in the following form:
[ ]
MD1(1),0,1
0
AD1(1),0,1
0
MTP(1(1), 0, 1) = .
UD1(1),0,1
1
MD 1(1),0,1
1
The parameter that is altered is the storage area capacity A = 0,1,2,3,…, while all
the rest parameters remain unchanged, i.e. M = 1 (supplier), P1 = 1 (parallel supply
channels in this supplier) and C = 1 channel at the DC. The matrix of the transition
probabilities for the system (1(1),1,1) has the following form presented in Table 2.
S1(1),1,1 = S1(1),0,1 + 1 = 3 + 1 = 4
while the system (1(1),A,1) has the form presented in Table 3.
So, the matrix of the transition probabilities of the examined system (1(1),A,1)
has the synoptic form:
[ ]
MD01(1),A,1 AD1(1),0,1
0
MTP(1(1), A, 1) = .
UD11(1),0,1 MD1(1),0,1
1
Summarizing, when the storage area capacity A increases, only the diagonal sub
matrix MD1(1),A,1
0
is affected. Specifically, as the storage area capacity increases, the
dimension of MD1(1),A,1
0
increases in exactly the same way. For the 1st additional
13
712 M. Vidalis et al.
1(1),A,1
⎛ S
⏞⏞⏞ ⎞
−4+(1−1)
⎜ ⎟
space the row ⎜𝜇, −𝜇1 − 𝜇, 𝜇1 , 0, … , 0 ⎟ is added, for the 2nd additional space
⎜ ⎟
⎝ ⎠
1(1),A,1
⎛⏞⏞
2−1
⏞
S
⏞⏞⏞ ⎞
−4+(2−1)
⎜ ⎟
the row ⎜ 0 , 𝜇, −𝜇1 − 𝜇, 𝜇1 , 0, … , 0 ⎟ is added,…, for the ith additional
⎜ ⎟
⎝ ⎠
1(1),A,1
⎛⏞⏞ i−1
⏞
S
⏞⏞⏞ ⎞
−4+(i−1)
⎜ ⎟
space the row with elements ⎜0, … , 0, 𝜇, −𝜇1 − 𝜇, 𝜇1 , 0, … , 0 ⎟ is added. The
⎜ ⎟
⎝ ⎠
rest sub matrices MD1(1),A,1
1
, AD1(1),A,1
0
and UD1(1),A,1
1
remain unchanged.
So, the dimension of the matrix of the transition probabilities or equivalently the
total number of states is given by the following relationship:
S1(1),A,1 = S1(1),0,1 + A
S1(1),A,1 = 3 + A
Inductive conclusion that results from the analysis above:
The parameter that is altered now is the capacity of the DC, C = 1,2,3,…, (channels)
while the rest parameters remain unchanged, i.e. M = 1 (suppliers), P1 = 1 (parallel
supply channels in this supplier) and A = 0.
13
Analysis of a two echelon supply chain with merging suppliers,… 713
The matrix of the transition probabilities for system (1(1),0,2) has the form
presented in Table 4, while the structure of the system (1(1),0,C) is presented in
Table 5.
The synoptic form of the system (1(1),0,C) is given by:
[ ]
MD01(1),A,1 AD1(1),0,1
0
MTP(1(1), A, 1) = .
C ∗ UD11(1),0,1 −C ∗ MD11(1),0,1
Consequently, the DC capacity C seems to affect mostly the diagonal sub matrix
MD1(1),0,1
0
. The dimension of MD1(1),0,C
0
increases correspondingly with the increas-
ing number of channels C, and especially in the same way. Any time we add a
new channel the MD1(1),0,C−1
0
increasing by one row and one column. Specifically,
the elements − μ1 and μ1 are inserted in the most upper cells i.e. (1,1) and (1,2)
and the element μ is inserted in cell (2,1). Finally, the synoptic form of the system
MTP(1(1),0,C − 1) gives the rest transition matrix for the examined case of capacity
C realizing two transitions: the symbols, − μ, and μ is added in all diagonal and sub-
diagonal cell values of MTP(1(1),0,C − 1) respectively.
So, as we can see from the finally synoptic form above the matrix-block AD1(1),0,C
0
remains unaltered compared with the matrix-block AD1(1),0,1
0
of the simplest system,
while the matrix-blocks UD1(1),0,C
1
and MD1(1),0,C
1
is almost the same with those of
the simplest system with the only difference that any element has multiplied with C.
The relationship that generates the dimension of the matrix of the transition prob-
abilities or equivalently the number of states is given below:
S1(1),0,C = S1(1),0,C−1 + 1, C = 1, 2, 3, … ,
13
714 M. Vidalis et al.
S1(1),0,C = S1(1),0,1 + (C − 1) = 3 + C − 1 = C + 2, C = 1, 2, 3, … ,
The inductive conclusion that comes out from the above analysis is:
This time the number of parallel supply channels P1 = 2,3,4,…, of the supplier is
altered, while all the other parameters remain unchanged, i.e. M = 1, A = 0 and C = 1.
Table 6 presents the structure of the matrix of the transition probabilities for system
(1(2),0,1):
13
Analysis of a two echelon supply chain with merging suppliers,… 715
As we can see from the above Tables when the capacity of the supplier is
increased the matrix of the transition probabilities MTP(1(P1),0,1) can be defined
from two structural procedures. The first one is that the matrix-block MD0 1
1(P ),0,1
comes from the transition matrix MTP(1(P1 − 1),0,1) by adding − μ1 to all diago-
nal elements and μ1 to all hyper-diagonal elements and the second is that just
after the matrix-block MD0 1 we add one column with element 𝜇1 (at the end
1(P ),0,1
of that column) and then one row with elements 𝜇 and −𝜇 (at the last two columns
of that row). We must mention here (this note will be used for the analysis below)
that if we have C capacity to the DC, the elements of the new row will be C ∗ 𝜇
and C ∗ (−𝜇) . The relationship that gives the number of states is the following:
In this case the number of suppliers (M) changes, while all the other parameters
remain stable. Initially, we examine the case with two suppliers that supply the
distribution center.
The possible states of that model are the following:
13
716 M. Vidalis et al.
From the above presentation the specific transition matrix can be represented as
follows:
[ ]
MD2(1,1),0,1
0
AD2(1,1),0,1
0
MTP(2(1, 1), 0, 1) = .
UD2(1,1),0,1
1
MD2(1,1),0,1
1
13
Analysis of a two echelon supply chain with merging suppliers,… 717
⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟
⎜
M 1, 1, … , 1⎟,0,1 ⎜
M−1 1, 1, … , 1⎟,0,1
⎜⏟⏞⏞⏟⏞⏞⏟⎟ ⎜⏟⏞⏞⏟⏞⏞⏟⎟
⎜ ⎟ ⎜ ⎟
S ⎝ M ⎠ =2⋅ S ⎝ M−1 ⎠ − 1, M = 2, 3, … ,
with primal term S 1(1),0,1
= 3 (recursive relationship of first order), or simply,
⎛ ⎞
⎜ ⎟
⎜
M 1, 1, … , 1⎟,0,1
⎜⏟⏞⏞⏟⏞⏞⏟⎟
⎜ ⎟
S ⎝ M ⎠ = 2M + 1, M = 2, 3, … ,
Secondly, the structure of the transition matrix can be summarized to the fol-
lowing four steps:
⎛ ⎞
⎜ ⎟
M ⎜1, 1, … , 1⎟,0,1
⎜⏟⏞⏞⏟⏞⏞⏟⎟ ⎛ ⎛ ⎞ ⎞
⎜ ⎟ ⎜ ⎜ ⎟ ⎟
⎝ ⎠
1. MD0 M
is a replica of MTP⎜M − 1⎜1, 1, … , 1⎟, 0, 1⎟ with three mod-
⎜ ⎜⏟⏞⏞⏟⏞⏞⏟ ⎟ ⎟
⎝ ⎝ M−1 ⎠ ⎠
ifications:
• μi has been replaced by μi+1 i = 1,2,…,M − 1, i.e. 𝜇i → 𝜇i+1
• −𝜇1 has been included to all diagonal cell values (in case there are P1 num-
ber of parallel supply channels
( )to the first supplier, it’s obvious that −𝜇1
will be substituted by P1 ∗ −𝜇1 ).
• 𝜇1 has been added to cell (1,2) (in case there are P1 number of parallel sup-
ply channels to the first supplier, 𝜇1 will be obviously substituted by P1 ∗ 𝜇1
and specifically, as we will show below, this will happen from cells (1,2),
(2,3), to cell (A + C,A + C + 1)).
⎛ ⎞ ⎛ ⎞
⎜ ⎟ ⎜ ⎟
M ⎜1, 1, … , 1⎟,0,1 M ⎜1, 1, … , 1⎟,0,1
⎜⏟⏞⏞⏟⏞⏞⏟⎟ ⎜⏟⏞⏞⏟⏞⏞⏟⎟
⎜ ⎟ ⎜ ⎟
⎝ ⎠ ⎝ ⎠
2. MD1 M
is a decreased reprint of MD0 M
with also three
⎛ ⎞
⎜ ⎟
M ⎜1, 1, … , 1⎟,0,1
⎜⏟⏞⏞⏟⏞⏞⏟⎟
⎜ ⎟
⎝ ⎠
modifications (the times that MD0 M
will
be repeated depends on the
number of parallel supply channels P1 , where in our case is one):
13
718 M. Vidalis et al.
⎛ ⎛ ⎞ ⎞
⎜ ⎜ ⎟ ⎟
The above analysis specifies that MTP⎜M ⎜1, 1, 1, … , 1⎟, 0, 1⎟ is based on
⎜ ⎜⏟⏞⏞⏞⏞⏟⏞⏞⏞⏞⏟⎟ ⎟
⎝ ⎝ M ⎠ ⎠
⎛ ⎛ ⎞ ⎞ ⎛ ⎛ ⎞ ⎞
⎜ ⎜ ⎟ ⎟ ⎜ ⎜ ⎟ ⎟
MTP⎜M − 1⎜1, 1, 1, … , 1⎟, 0, 1⎟ , and MTP⎜M − 1⎜1, 1, 1, … , 1⎟, 0, 1⎟ is based on
⎜ ⎜⏟⏞⏞⏞⏞⏟⏞⏞⏞⏞⏟⎟ ⎟ ⎜ ⎜⏟⏞⏞⏞⏞⏟⏞⏞⏞⏞⏟⎟ ⎟
⎝ ⎝ M−1 ⎠ ⎠ ⎝ ⎝ M−1 ⎠ ⎠
⎛ ⎛ ⎞ ⎞ ⎛ ⎛ ⎞ ⎞
⎜ ⎜ ⎟ ⎟ ⎜ ⎜ ⎟ ⎟
MTP⎜M − 2⎜1, 1, 1, … , 1⎟, 0, 1⎟ and so on. Consequently, MTP⎜M ⎜1, 1, 1, … , 1⎟, 0, 1⎟
⎜ ⎜⏟⏞⏞⏞⏞⏟⏞⏞⏞⏞⏟ ⎟ ⎟ ⎜ ⎜ ⏟⏞⏞⏞⏞⏟⏞⏞⏞⏞⏟ ⎟ ⎟
⎝ ⎝ M−2 ⎠ ⎠ ⎝ ⎝ M ⎠ ⎠
is based on MTP(1(1), 0, 1).
In all the above procedure, the impact of each parameter A, C, Pj j = 1,…, M, and
M in isolation has been examined. The jointed impact of all parameters is examined
below.
13
Analysis of a two echelon supply chain with merging suppliers,… 719
In this section the way that the transition matrix is generated when all the system
parameters A, C, Pj , and M, change at the same time is described. The procedure
uses as a seed the fundamental matrix (i.e. the transition matrix of the simplest
model (1(1),0,1)) described in Sect. 3. In order to illustrate the generation of the
matrix of the transition probabilities an example is used. More specifically it is
shown how the matrix of the transition probabilities of the system (3(2,3,4),1,2) can
be generated using only as a basis the matrix of the transition probabilities of system
(1(1),0,1)). The matrix of the transition probabilities of system (3(2,3,4),1,2), due to
his huge dimension, 63 × 63, is presented in a compact form (see Table 10), which
has been introduced in the previous sections. The complete matrix of the transition
probabilities in an analytical form is given in “Appendices A and B”.
The generation of the matrix of the transition probabilities can be done only
when the following sequence A → C → Pj → M of the considered parame-
ters is applied. Therefore firstly (according to the characteristics of the system
(3(2,3,4),1,2)), the storage area capacity is considered, and in comparison with
the fundamental matrix A has been increased from zero to one ( A = 1 ). Since
the storage area capacity has been changed the procedure described in Sect. 3.1
can be applied in order to get the matrix of the transition probabilities presented
in Table 5 which corresponds to system (1(1),1,1). Starting now from system
(1(1),1,1) the DC capacity is increased by one (i.e. C = 2 ). Applying the meth-
odology that was presented in Sect. 3.2, the matrix of the transition probabilities
presented in Table 11 is obtained which corresponds to system (1(1),1,2).
Using now system (1(1),1,2) as a starting point and considering the target sys-
tem (3(2,3,4),1,2)), the procedure takes into account the number of parallel sup-
ply channels of the supplier with the largest index (i.e., the third supplier which
has four parallel supply channels) and continuous examining suppliers in decreas-
ing order of their index. Therefore, the system that needs to be examined is the
system (1(4),1,2). Applying the procedure described in paragraph 3.3 the matrix
of the transition probabilities presented in Table 12 is obtained.
Starting now from system (1(4),1,2) the next step is to generate the matrix of
the transition probabilities of the system (2(3,4),1,2) (see “Appendix B”), adding
13
720 M. Vidalis et al.
the second supplier with capacity 3. The applied steps are the same with the ones
described in Sect. 3.4 with a slight modification. In particular, step 1 is exactly
the same, while in step 2 except from the row and column that should be removed
it is necessary to remove the rows and columns that correspond to the capacity
A of the storage area, and capacity C at the DC, namely three rows and columns.
Furthermore, step 2 should be implemented P1 times and each time one more 𝜇1
should be removed from all diagonal elements (totally at the last implementa-
tion P1 elements should be removed). As long as step 2 is performed, steps 3
and 4 should be implemented simultaneously. Step 4 remains the same, creating
a square matrix with all the diagonal elements equal to 2 ∗ 𝜇 (since C = 2 ), with
dimension similar to the matrix-block that is created from step 2, while in step 3 a
square matrix with all diagonal elements equal to 3 ∗ 𝜇1 (since P1 = 3 ) is created.
For each sequential application of step 2 𝜇1 should be subtracted from all diago-
nal elements of the matrix created by step 3.
Finally, the matrix of the transition probabilities of the system (2(3,4),1,2) can
be used in order to create the matrix of the transition probabilities of the system
(3(2,3,4),1,2) by adding one supplier at the beginning with capacity 2 and apply-
ing the steps that described in Sect. 3.4 with the modifications that were men-
tioned before.
Generally, the matrix of the transition probabilities of any system can be created
by changing one parameter at a time starting with A and ending with M following
the sequence A → C → Pj → M , using the corresponding methodology exactly with
the same manner for all parameters except M where there are slight modifications
that are compatible with the common sense. Therefore if the matrix of the transition
probabilities of the system (4(3,2,5,7),3,4) needs to be created the following matri-
ces of the transition probabilities of the corresponding systems should be created
using the sequence that is presented below:
(1(1), 0, 1) → (1(1), 3, 1) → (1(1), 3, 4) → (1(7), 3, 4)
→ (2(5, 7), 3, 4) → (3(2, 5, 7), 3, 4) → (4(3, 2, 5, 7), 3, 4)
4 The algorithm
13
Analysis of a two echelon supply chain with merging suppliers,… 721
5 Performance measures
Using the steady-state probabilities calculated from the above procedures we can
estimate all the performance measures of any examined system. The performance
measures, which are more important, are the average invetory in the DC (WIPDC),
the average invetory of the system (WIPsystem), and the throughput or the mean out-
put rate of the system (THR).
The first performance measure can be defined as:
WIPDC = Mean number of Blocked Parallel Supply channels
+ Mean number of busy channels in DC + Mean Storage area Level
13
722 M. Vidalis et al.
⎡ P1 + P2 + ⋯ + PM , P1 + P2 + ⋯ + PM + 1, P1 + P2 + ⋯ + PM + 2, … , ⎤
⎢ (P1 +2P2 +3P3 +⋯+MPM )times ⎥
z=⎢ ⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞ ⎥
⎢ P + P + ⋯ + P + C + A, P + P + ⋯ + P + C + A ⎥
⎣ 1 2 M 1 2 M ⎦
where the last element of the vector z denotes the times that we have in total
P1 + P2 + ⋯ + PM + A + C components in the system regarding only the cases
where we have one or more blocked parallel supply channels.
�
P1 +P2 +⋯+PM +C+A
WIPsystem = k ∗ 𝜋11 … 1 11 … 1,…,11 … 1 (k−(P1 +P2 +⋯+PM ))
k=P1 +P2 +⋯+PM ��� ��� ���
P1 P2 PM
� �
+ P1 + P2 + ⋯ + PM + C + A
⎛ ⎞
⎜ ⎟
⎜ ⎛ ⎞⎟
⎜ ⎜ ⎟⎟⎟
⎜ ⎜ ⎟⎟
⎜� P1 �P2 �P M
⎜ ⎟⎟
∗⎜ ⋯ ⎜ 𝜋 ⎟⎟, for i + j + ⋯ + n > 0.
⎜ i=0 j=0 n=0 ⎜ nTimes
��� ⎟⎟
⎜ ⎜ 11 … b
⎜
iTimes
11 … b jTimes
,…,11 … b (C+A) ⎟⎟
⎜ ⎜ ����� � ����� �����
� ����
� ������ � ������ ⎟⎟
⎝ ⎠
⎜��������������������������������������������������������������������������������������������������� ⎟
P1 P2 PM
⎜ ⎟
⎝ (P1 +2P2 +3P3 +⋯+MPM )probabilities ⎠
The third performance measure i.e. the mean output rate or the throughput is
given by the following relationship:
THR = μ · Pr[DC is busy]. The analytical form of the throughput is
13
Analysis of a two echelon supply chain with merging suppliers,… 723
⎛ ⎞
�
C−1
⎜ ⎟
THR = j ∗ 𝜇 ∗ ⎜𝜋11 … 1 11 … 1,…,11 … 1 j ⎟
j=1 ⎜ ⏟⏟⏟ ⏟⏟⏟ ⏟⏟⏟ ⎟
⎝ P1 P2 PM ⎠
⎛ ⎛ ⎞⎞
⎜ ⎜ �
C−1
⎟⎟
+⎜(C ∗ 𝜇) ∗ ⎜1 − 𝜋11 … 1 11 … 1,…,11 … 1 j ⎟⎟
⎜ ⎜ j=1 ⏟⏟⏟ ⏟⏟⏟ ⏟⏟⏟ ⎟⎟
⎝ ⎝ P1 P2 PM ⎠⎠
In order to validate the proposed method, the results of analytical model were com-
pared against those obtained from simulation. For this reason the considered model
was simulated in Arena.
The simulation runs used one replication of 250 million time units and 95% con-
fidence intervals for all performance measures with half-width values of less than
0.07% of their respective estimated values. The numerical results showed a very
good agreement with simulation. The percentage error comparing with simulation
for all cases lies between − 0.14 and 0.09%.
Additionally to the existing numerical results presented in the previous section that
showed a very good agreement between the proposed algorithm and simulation,
this section explores the effect of the system variables on the system’s performance
measures.
The proposed model is used to investigate the effect of each parameter M, Pj , 𝜇j ,
C, μ, and A on the performance measures of the system. Any system performance
measure can be evaluated using the above algorithm. Despite this, due to space limi-
tations numerical results only for some of the above measures such as throughput
and WIPsystem are presented in this section. The considered system is examined under
two different cases. The first case considers balanced systems, i.e. systems where the
potential total inbound rate to the storage area equals the total outbound rate, namely
∑M
P ⋅ 𝜇j = C ⋅ 𝜇 = 1 . The second case considers constrained systems, i.e. sys-
j=1 j
tems where the mean processing rate for all merging suppliers is less than the mean
∑M
processing rate for all machines at the DC, namely j=1 Pj ∗ 𝜇j = 1 < C ∗ 𝜇 = 2.
For each one of the above cases three different scenarios are associated. The first
“homogeneous” scenario included systems where each one of the M merging sup-
pliers has the same rate, μi = 1/M, i = 1,2,…, M. The second “ranked heterogene-
ous” scenario includes systems where the first supplier has the smallest processing
rate and the processing rate of the rest suppliers increases as their corresponding
index increase (i.e., 𝜇1 < 𝜇2 < ⋯ < 𝜇M ). The last “random heterogeneous” scenario
includes systems where the processing rates are randomly distributed among the
suppliers.
The data used for the “ranked heterogeneous” and “random heterogeneous” sce-
narios are given in Table 13.
13
724 M. Vidalis et al.
The balanced case and especially the evolution of throughput for the above three
scenarios for different combinations of system parameters is examined first. More
specifically, for each number of merging suppliers M = 1,…,5, the case where
1 ≤ Pj ≤ 10, A = 5 and C = 1 and the case where 1 ≤ Pj ≤ 10, A = 5 and C = 5 are con-
sidered. The shape of the surfaces (concave and smooth), regarding the evolution
of throughput indicates that there is always a unique maximum value of throughput
which is obtained for a specific combination of the balanced system’s parameters.
Due to space limitations the curve of throughput for homogeneous and random het-
erogeneous configurations for M = 2 suppliers, 1 ≤ Pj ≤ 10, A = 5 and C = 5 is pre-
sented in Fig. 2.
The next point of interest is the search of the optimal vector of supplier’s, Pj ,
j = 1,2,…,M, storage area and DC capacity that maximizes the throughput of sys-
tems with M = 2, 3, 4 and 5 suppliers, 1 ≤ Pj ≤ 5, 0 ≤ A ≤ 10 and 1 ≤ C ≤ 5, for the
balanced case.
The numerical results given in Table 14, indicate that for each scenario and each
value of M merging suppliers, the objective to maximize the throughput can be
achieved by selecting the maximum feasible values of the system parameters (i.e.,
the maximum values of Pj, A and C). Another rule that comes out from these results
is that while the number of M merging suppliers increases, the throughput of the
system in all scenarios increases (even with a slight difference) as well. Therefore
the most critical point to get the maximum throughput value is to use the maximum
feasible values of the rest system parameters. Finally it can be noticed that for the
same number M of suppliers all three scenarios provide almost the same throughput
value when the maximum values of Pj, A and C are used.
The evolution of the performance measure WIPsystem, for the balanced case and
for all three different scenarios is examined next. More specifically, for each num-
ber of merging suppliers M = 1,…,5, the case where 1 ≤ Pj ≤ 10, A = 5 and C = 1
and the case where 1 ≤ Pj ≤ 10, A = 5 and C = 5 are considered. Like in case where
the throughput was examined, the resulting shapes indicate that there is always a
minimum value of WIPsystem which we can be obtained for specific combinations
of the system’s parameters. Due to space limitations only the curves of W IPsystem
for the homogeneous and the random heterogeneous scenarios for M = 2 suppliers,
1 ≤ Pj ≤ 10, A = 5 and C = 5 are presented in Fig. 3.
Table 13 Data used for the “ranked heterogeneous” and “random heterogeneous” scenarios
Number of Processing rates Processing rates
suppliers M Ranked Heterogeneous scenario Random Heterogeneous scenario
13
Analysis of a two echelon supply chain with merging suppliers,… 725
P1
P2 P1
P2
Fig. 2 Balanced Case: Evolution of Throughput for Homogeneous (left) and random heterogeneous
(right) configurations for M = 2 suppliers, 1 ≤ Pj ≤ 10, A = 5 and C = 5
Like in the case of throughput, the optimal vector of supplier’s, Pj , j = 1,2,…,M,
storage area capacity, A and DC capacity, C that minimizes WIPsystem for systems
with M = 2, 3, 4 and 5 suppliers, 1 ≤ Pj ≤ 5, 0 ≤ A ≤ 10 and 1 ≤ C ≤ 5, for the all three
different scenarios is presented in Table 15.
The results presented in Table 15 indicate that for each scenario and each value
of M, M = 2,…,5 merging suppliers, the minimization of WIPsystem is obtained by
selecting the minimum feasible values of each system parameter (i.e., the minimum
values of Pj, A and C). Moreover, while the number of merging suppliers decreases,
the WIPsystem decreases as well for all the scenarios examined. Despite that, for the
same number of merging suppliers M, all the scenarios provide almost the same
minimum values of WIPsystem, the scenario of heterogeneous systems provide always
slightly lower WIPsystem values than the homogeneous scenario.
The next optimization problem examined in this section is to find the optimal
combination of the system parameters (i.e., the optimal vector of the system param-
eters) that achieves at least a desired level of throughput, THRlevel and at the same
time obtain the minimum feasible work in process of the system. The values of the
system parameters for the numerical results given in Table 16 for all scenarios are
M = 2, 3 4, 1 ≤ Pj ≤ 5, 0 ≤ A ≤ 15 and 1 ≤ C ≤ 5.
The numerical results presented in Table 16, indicate that for every scenario
examined, the system that achieves a specific desired throughput level and at the
same time provides the minimum feasible WIPsystem should have the minimum
13
726 M. Vidalis et al.
P1
P2
P1 P2
Fig. 3 Evolution of W IPsystem for homogeneous (left) and random heterogeneous (right) scenario for
M = 2 suppliers, 1 ≤ Pj ≤ 10, A = 5 and C = 5
number of parallel supply channels at each supplier (i.e., one supply channel at
each supplier) and the minimum number of identical machines at the DC (i.e., one
machine at the DC). Additionally, for a fixed number of merging suppliers, the
capacity of the shared storage area increases when the desired throughput level
increases as well. Finally, the heterogeneous scenarios provide lower W
IPsystem val-
ues than the homogeneous scenario and more specifically the ranked heterogeneous
scenario provides the minimum WIPsystem values among all the scenarios examined.
This section presents a similar analysis like the one given in the previous section for
the case of constrained systems under the three scenarios. The shape of the specific
surfaces (concave and smooth) for the constrained case have the same pattern as in
the unconstrained case regarding the evolution of throughput and WIPsystem. These
shapes indicate that there is always a unique maximum value of throughput and a
unique minimum value of WIPsystem which is obtained for a specific combination of
the constrained system’s parameters. Due to space limitations these surfaces are not
presented in this section. The evolution of throughput and W IPsystem is examined for
all scenarios and combinations of the system parameters that were used for the bal-
anced case.
13
Analysis of a two echelon supply chain with merging suppliers,… 727
Table 16 Optimal combination of the system parameters to achieve at least a desired level of throughput
and obtain the minimum feasible WIPsystem
M THRlevel minWIPsystem minWIPsystem minWIPsystem Ran- Optimal vector
Ηomogeneous Ranked dom Ηeterogeneous for all configurations
scenario Ηeterogeneous scenario (P1,…,PM,A,C)
scenario
The search of the optimal vector of the system parameters that maximizes through-
put or minimize W IPsystem for systems with M = 2, 3, 4 and 5 suppliers, 1 ≤ Pj ≤ 5,
0 ≤ A ≤ 10 and 1 ≤ C ≤ 5, for the three constrained scenarios is considered first.
Table 17 provides similar results with the ones obtained for the balanced case.
More specifically, for all scenarios, the maximum value of throughput or the mini-
mum value of W IPsystem is achieved when the system uses the maximum or the
minimum feasible values of the system parameters, respectively. Furthermore, for
all scenarios and for every number of M merging suppliers the values of maxTHR
are exactly the same. Additionally, when M decreases the m inWIPsystem also
decreases, which is reasonable. Finally, for each number of merging suppliers M, the
minWIPsystem value is observed to the Ranked Ηeterogeneous scenario.
The last point of interest, is to find the optimal vector that achieves at least a
desired level of throughput, and at the same time provides the minimum feasible
work in process of the system, WIPsystem. The same values of the system parameters
like the ones used for the balanced case are used here.
From the above table it is clear that for every scenario all the system parameters
should get their minimum values. In contradiction with the balanced case for most
of the examined cases the storage area capacity is zero and only when the desired
throughput level is very high the storage area capacity becomes equal to one (which
is still the minimum positive value). This is reasonable since the value of the mean
processing rate of the DC is two times the value of the mean processing rate of the
merging suppliers. For the same reason all scenarios have almost the same perfor-
mance measures for each combination of the system parameters.
13
728 M. Vidalis et al.
Summarizing all the results presented in Tables 14, 15, 16, 17 and 18 and in
Figs. 2 and 3 the following conclusions can be done:
A two echelon merge supply network is analyzed in this paper. The considered sys-
tem was modelled as continuous time discrete state Markov chain and an algorithm
for its performance evaluation was also presented. The numerical results from the
proposed algorithm showed a very good agreement with simulation. Additionally
the impact of the system parameters on its performance measures was also explored
using as a tool the proposed evaluative algorithm. The contribution of this paper is
that the proposed method can be used for the performance evaluation and optimi-
zation of merge-assembly supply chains and help practitioners to make decisions
about the optimal design and parameter’s selection in order to reduce operational
inefficiencies.
13
Table 17 Optimal vectors of the constrained system to maximize throughput or minimize WIPsystem
M Ηomogeneous Optimal maxTHR/minWIPsystem Ranked Ηeterogeneous maxTHR/minWIPsystem Random maxTHR/minWIPsystem
vector (P1,…,PM,A,C) Optimal vector Ηeterogeneous Optimal
maxTHR/minWIPsys (P1,…,PM,A,C) vector (P1,…,PM,A,C)
maxTHR/minWIPsystem maxTHR/minWIPsystem
13
730 M. Vidalis et al.
Table 18 Optimal combination of the system parameters in order to obtain the minimum feasible value
of WIPsystem achieving at least a desired level of throughput (constrained case)
M THRlevel minWIPsystem minWIPsystem minWIPsystem Optimal vector
Ηomogeneous Ranked Random for all configurations
Ηeterogeneous Ηeterogeneous (P1,…,PM,A,C)
An extension of this research could be the analysis of a similar system where the
suppliers or unreliable machines at the DC may be unreliable. Another extension
could be the use of phase type distributions to model the processing rates of the sup-
pliers and of the machines at the DC. Finally the examined system may be consid-
ered as a push–pull type system with the addition of a storage area of finished goods
(BFG) after the DC that will satisfy an external demand rate.
Appendix A
Furthermore:
The matrix of the transition probabilities of the sub matrix (−S1 ∗𝜇1 ) D3(2,3,4),1,2
1
is the
same with the sub matrix (−1∗𝜇1 ) D3(2,3,4),1,2
1
, with the only difference that −𝜇 1 has
been added to all diagonal cell values except from the last one.
The matrix of the transition probabilities of the sub matrix S1 ∗ U03(2,3,4),1,2 is a
diagonal matrix with all diagonal cell values equal to 2 ∗ 𝜇1 (since S1 = 2).
The matrix of the transition probabilities of the sub matrix (S1 − 1) ∗ U03(2,3,4),1,2 is
a diagonal matrix with all diagonal cell values equal to 1 ∗ 𝜇1 (since S1 − 1 = 1).
The matrix of the transition probabilities of the sub matrix N ∗ L13(2,3,4),1,2 is a
diagonal matrix with all diagonal cell values equal to 2 ∗ 𝜇1 (since N = 2 )
(see Tables 19, 20).
13
0
Table 19 Matrix of the transition probabilities of sub matrix D3(2,3,4),1,2
2340 2341 2342 2343 2333 2323 2313 23b3 2243 2233 2223 2213
13
Table 19 (continued)
732
2340 2341 2342 2343 2333 2323 2313 23b3 2243 2233 2223 2213
21b3
13
2b43
2b33
2b23
2b13
2bb3
22b3 2143 2133 2123 2113 21b3 2b43 2b33 2b23 2b13 2bb3
2340
2341
2342
2343
2333
2323
2313
23b3 3μ2
2243 2μ2
2233 2μ2
2223 2μ2
2213 μ3 2μ2
22b3 − 2μ1 − 2μ2 2μ2
− 2μ
2143 − 2μ1 − μ2 4μ3 μ2
− 4μ3 − 2μ
2133 − 2μ1 − μ2 3μ3 μ2
− 3μ3 − 2μ
M. Vidalis et al.
Table 19 (continued)
22b3 2143 2133 2123 2113 21b3 2b43 2b33 2b23 2b13 2bb3
− 2μ
733
13
1
Table 20 Matrix of the transition probabilities of sub matrix (−1∗𝜇1 ) D3(2,3,4),1,2
734
1343 1333 1323 1313 13b3 1243 1233 1223 1213 12b3
13
1343 − μ1 − 3μ2 4μ3 3μ2
− 4μ3 − 2μ
1333 − μ1 − 3μ2 3μ3 3μ2
− 3μ3 − 2μ
1323 − μ1 − 3μ2 2μ3 3μ2
− 2μ3 − 2μ
1313 − μ1 − 3μ2 μ3 3μ2
− μ3 − 2μ
13b3 − μ1 − 3μ2 − 2μ 3μ2
1243 − μ1 − 2μ2 − 4μ3 − 2μ 4μ3
1233 − μ1 − 2μ2 − 3μ3 − 2μ 3μ3
1223 − μ1 − 2μ2 2μ3
− 2μ3 − 2μ
1213 − μ1 − 2μ2 μ3
− μ3 − 2μ
12b3 − μ1 − 2μ2 − 2μ
1143
1133
1123
1113
11b3
1b43
1b33
1b23
1b13
1bb3
M. Vidalis et al.
Table 20 (continued)
1143 1133 1123 1113 11b3 1b43 1b33 1b23 1b13 1bb3
1343
1333
1323
1313
13b3
1243 2μ2
1233 2μ2
1223 2μ2
1213 2μ2
12b3 2μ2
1143 − μ1 − μ2 4μ3 μ2
− 4μ3 − 2μ
1133 − μ1 − μ2 3μ3 μ2
− 3μ3 − 2μ
Analysis of a two echelon supply chain with merging suppliers,…
13
Appendix B
736
13
Table 21 Matrix of the transition probabilities of sub matrix (2(3,4),1,2)
340 341 342 343 333 323 313 3b3 243 233 223 213
b33
b23
b13
bb3
2b3 143 133 123 113 1b3 b43 b33 b23 b13 bb3
340
341
342
343
333
323
313
3b3 3μ1
Analysis of a two echelon supply chain with merging suppliers,…
243 2μ1
233 2μ1
223 2μ1
213 μ2 2μ1
2b3 − 2μ1 − 2μ 2μ1
143 − μ1 − 4μ2 − 2μ 4μ2 μ1
133 − μ1 − 3μ2 − 2μ 3μ2 μ1
123 − μ1 − 2μ2 − 2μ 2μ2 μ1
113 − μ1 − μ2 − 2μ μ2 μ1
1b3 2μ − μ1 − 2μ μ1
737
13
Table 21 (continued)
738
2b3 143 133 123 113 1b3 b43 b33 b23 b13 bb3
13
b43 2μ − 4μ2 − 2μ 4μ2
b33 2μ − 3μ2 − 2μ 3μ2
b23 2μ − 2μ2 − 2μ 2μ2
b13 2μ − μ2 − 2μ μ2
bb3 2μ − 2μ
M. Vidalis et al.
Analysis of a two echelon supply chain with merging suppliers,… 739
References
Baumann H, Sandmann W (2017) Multi-server tandem queue with Markovian arrival process, phase-type
service times and finite buffers. Eur J Oper Res 256:187–195
Burman MH (1995) New results in flow line analysis. Ph.D. Thesis, Operations Research Center, Mas-
sachusetts Institute of Technology, USA
Chen LG, Ding D, Ou J (2014) Power structure and profitability in assembly supply chains. Prod Oper
Manag 23(9):1599–1616
Diamantidis AC, Papadopoulos CT, Vidalis MI (2004) Exact analysis of a discrete material three station
one buffer merge system with unreliable machines. Int J Prod Res 42(4):651–675
Diamantidis AC, Papadopoulos CT, Heavey C (2007) Approximate analysis of serial flow lines with mul-
tiple parallel-machine stations. IIE Trans 39(4):361–375
Diamantidis AC, Koukoumialos S, Vidalis M (2016) Performance Evaluation of a push-pull merge sys-
tem with multiple suppliers, an intermediate buffer and a distribution centre with parallel machines/
channels. Int J Prod Res 54(9):2628–2652
Dijk N, Wal J (1989) Simple bounds and monotonicity results for finite multi-server exponential tandem
queues. Queuing Syst 4(1):1–16
Dubois D, Forestier JP (1982) Productivité et en-cours moyens d’un ensemble de deux machines se
parées par une zone de stockage. RAIRO Automat 16(2):105–132
Forestier JP (1980) Modélisation stochastique et comportement asymptotique d’un système automatise de
production. RAIRO Automat 14(2):127–143
Friedman HD (1965) Reduction methods for tandem queuing systems. Oper Res 13(1):121–131
Geranios M, Vidalis M, Vrisagotis V (2016) Evaluating the performance of a merge supply network with
suppliers interruptions. Int J Decis Support Syst 2:91–108
Goldratt EM (1990) Theory of constraints. North River Press, Great Barrington
Hamta N, Shirazi MA, Behdad S, Ghomi SMTF (2018) Modeling and measuring the structural complex-
ity in assembly supply chain networks. J Intell Manuf 29(2):259–275
Helber S, Mehrtens N (2003) Exact analysis of a continuous material merge system with limited buffer
capacity and three stations. In: Gershwin SB, Dallery Y, Papadopoulos CT, MacGregor Smith J
(eds) Analysis and modeling of manufacturing systems (Kluwer), pp 85–121
Iyama T, Ito S (1987) The maximum production rate for an unbalanced multi-server flow line with finite
buffer storage. Int J Prod Res 25(8):1157–1170
Li J, Blumenfeld Dennis, Huang Ningjian, Alden Jeffrey (2009) Throughput analysis of production sys-
tems: recent advances and future topics. Int J Prod Res 47(14):3823–3851
Liu Y, Li J (2009) Modeling and analysis of split and merge production systems with Bernoulli reliability
machines. Int J Prod Res 47(16):4373–4397
Liu Y, Li J (2010) Split and merge production systems: performance analysis and structural properties.
IIE Trans 42(6):422–434
Modrak V, Marton D (2013) structural complexity of assembly supply chains: a theoretical framework.
Procedia CIRP 7:43–48
Patchong A, Willaeys D (2001) Modeling and analysis of an unreliable flow line composed of parallel-
machine stages. IIE Trans 33(7):559–568
Quazene Y, Yalaoui A, Chehade H, Yalaoui F (2013) A buffered three station merge system: performance
analysis and comparative study. In: International conference on control, decision and information
technologies, CoDIT, pp 154–159
Shin YW, Moon DH (2014) Approximation of throughput in tandem queues with multiple servers and
blocking. Appl Math Model 38(24):6122–6132
Shin YW, Moon DH (2017) Throughput of flow lines with unreliable parallel-machine workstations and
blocking. J Ind Manag Optim 13(2):901–916
Tan B (2001) A three station merge system with unreliable stations and a shared buffer. Math Comput
Model 33:1011–1026
Vidalis M, Koukoumialos S, Dio DD, Varlas G (2012) Performance evaluation of a merge supply system
with a distribution centre, two reliable suppliers, one buffer and Erlang lead times. Int J Bus Sci
Appl Manag 7(3):42–55
Vidalis MI, Koukoumialos St, Geranios M (2014) Performance evaluation of a merge supply network: a
distribution centre with multiple reliable random suppliers. Comput Ind Eng 70:43–58
13
740 M. Vidalis et al.
Vidalis MI, Koukoumialos St, Diamantidis AC, Blanas G (2015) Performance evaluation of a merge sup-
ply network: a production centre with multiple different reliable suppliers. In: 10th Conference on
Stochastic models of manufacturing and service operations (SMMSO), Volos, Greece, pp 255–263
Wu CH, Lee WC, Ke JC, Liu TH (2014) Optimization analysis of an unreliable multi-server queue with a
controllable repair policy. Comput Oper Res 49:83–96
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.
13