Strategic DDMRP's Buffer Positioning For Hybrid MTO-MTS Manufacturing

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

2020 IEEE International Conference on Technology Management, Operations and Decisions (ICTMOD) | 978-1-7281-5950-8/20/$31.00 ©2020 IEEE | DOI: 10.1109/ICTMOD49425.2020.

9380588

Strategic DDMRP’s Buffer Positioning for hybrid MTO/MTS manufacturing

Abdelhalim Achergui, Hamid Allaoui,Tiente Hsu


Univ. Artois, UR 3926, Laboratoire de Génie
Informatique et d’Automatique de l’Artois (LGI2A),
62400 Béthune, France
Email: (abdelhalim.achergui,hamid.allaoui,tiente.hsu)@univ-artois.fr

Abstract—To make a product, many components: made Demand Driven Material Requirements Planning
inside a supply chain or provided by external suppliers, may (DDMRP), C. Ptak, C. Smith [1] was a result of combining
be needed as specified in the bill of material (BOM). This the best practices of many other methods such as MRP,
variety exposes the production process to various risks as
stock ruptures for instance. Due to companies efforts to face Kanban and JIT. This method is based on decoupling
the volatile demand and market pressure, Work-in-process points, decomposing the flow line into short parts avoiding
(WIP) inventories become redundant in flow production lines. the bullwhip effect in the supply chain. Each of these
Uncertainty became a major flaw of the system, to which many points is equipped with a buffer, serving as a visual
methods were developed and adopted to manage and optimize indicator representing the net flow equation, that we
the use of resources. One of the most recent methods is Demand
Driven Material Requirement Planning (DDMRP) that uses explain later in this article. It has three zones (Red, Yellow
buffers: visual indicators of the net flow integrating the real and green) indicating each the situation of the flow of
stock, the actual demand and the expected provisions; This materials and triggering replenishment orders based on real
paper describes an approach to solve the optimization problem demand. The method suggests a qualitative positioning
of minimizing the storing costs, under service time constraint strategy, depending on factors related to the client, the
for uncapacitated buffer positioning alongside with the buffer
allocation following the DDMRP planning methodology in a market and some other production specifics. The intent of
hybrid MTO/MTS manufacturing system. Finally, numerical this paper is to optimize buffer positioning for a hybrid
examples are applied to test the efficiency and the performance MTO/MTS manufacturing flow line as did X. Y. Sun et
of the developed algorithm. al. [2], simultaneously with a buffer allocation, all while
Keywords-Buffer allocation problem, Buffer positioning maintaining a DDMRP logic.
problem, Make-to-order, Make-to-stock, DDMRP. The remainder of this paper is organized as follows: In
the next section, the relevant literature about the problem is
I. I NTRODUCTION reviewed and briefly discussed. In Section 3, an overview
O wadays the industrial revolution, production knew of the DDMRP methodology is given with a problem
N major development in its supply chain management.
Product became more complex and expensive. Client be-
statement for the work presented in this article. In Section
4, the proposed solving method is formulated and examined.
came pickier and more impatient. The market is now Numerical tests and performance evaluations are presented
competitive for a better service, “as soon as possible” and analyzed in Section 5. Finally, Section 6 concludes
delivery and for a low-cost configuration. As a result, the the paper and sums some perspectives for future research
manufacturers need to adapt their entire supply network, directions.
speed up their productivity and optimize their resources
in the flow lines. Research has been conducted to better II. L ITERATURE REVIEW
manage and track the production processes, and also to Many papers handled the Buffer Allocation Problem
optimize the throughput, the costs and the production lead (BAP) before in the literature, but few did both buffer
times. A major complication to do so, is the uncertain and allocation and buffer positioning. S. Weiss et al. [3] cate-
non-stationary demand phases of a product life cycle. With gorized in their review the existing work by three criteria:
these fluctuation, producers had several losses between the the characteristics of the supply chain, the objective function
inventory shortages and the unsold unsuccessful product and constraints and the solution methodology. Whereas G.
series. The resulting problems can be mitigated by putting up Liberopoulos [4] found that any work ever done was in the
inventories. However, having more inventories than needed purpose of one of these objectives: modeling, performance
can be costly. The decision on where to install buffers for evaluation or optimization. The first assumptions for a multi-
decoupled products along the Bill of materials (BOM) is echelon stock placement, which is another appellation for
known as the Buffer positioning problem. Different varia- the variety of stages in a bill of materials, were described
tions of this problem were discussed in the literature. by K. F. Simpson [5] and generalized by S. C. Graves

Authorized licensed use limited to: University of Canberra. Downloaded on May 20,2021 at 23:25:11 UTC from IEEE Xplore. Restrictions apply.
and S. P. Willems [6]. The first attempt of optimization the most optimized manufacturing options. In this context,
though, was done by E. Koenigsberg [7], where he tried DDMRP came with a model to absorb and control the
to discuss how to decrease the loss transfer, the stochastic suppliers and demand variability. Canceling as a result the
aspect and the limitation on the size of queues. It was defined bullwhip effect affecting the strategic and tactical plans of
afterwards as an NP-hard optimization problem by J. M. the manufacturer. The objective being to support and protect
Smith and F. R. B. Cruz [8]. According to the state of the flow of materials and information inside the supply
the art, we found that many methods were used to solve network. C. Ptak, C. Smith [1].
this problem. The following works used simulation : P.
Glasserman and S. Tayur [9] combined it with variation B. Decoupling Point (DP)
analysis to solve the Base-stock allocation problem. D. C. Defined by the dictionary of the American Production
Whybark and S. Yang [10] used it to get the best service and Inventory Control Society (APICS) as: “Creating in-
level given a set of resources. Y. Park et al. [11] did a dependence between supply and use of material. Commonly
simulation-based case study on a tv manufacturing plan. denotes providing inventory between operations so that fluc-
M. Amiri and A. Mohtashami [12] used the simulation to tuations in the production rate of the supplying operation do
solve the problem in an unreliable production environment. not constrain production or use rates of the next operation.
L. Tiacci [13] also examined the problem by simulation (p. 43)” The concept was introduced in the supply chain
techniques in production line with stochastic task times. A. context by G. Sharman [26]. His idea was that a DP is
Costa et al. [14] used a linear complex integer programming a compromise between the competitive demand and the
to simulate an adaptive local Parallel Tabu Search to solve product cost and complexity. And to hold that bargain, these
the primary problem. S. Y. Kose and O. Kilincci [15] based physical points must provide an absorption of the variability
their optimization on the simulation approach through a of both demand and supply at the same time. Which is
meta-heuristic algorithm combining genetic algorithm and made possible by decoupling inventories, or as referred to
simulated annealing. Also, dynamic programming has been in the DDMRP method by decoupling point buffer. They
used to optimize safety stock placement by the following are a quantity of inventory held at the decoupling point to
works : in the guaranteed-service model introduced by Simp- guarantee a reliable product availability while transmitting a
son [5], inventory placement problem for a serial process realistic and a non-distorted signal to the upstream stations
is successfully converted into allocation problem of service and suppliers. The main purposes of these buffers, which are
time to stages in the process. S. C. Graves [16] developed the heart of a DDMRP system and its visual management
later an optimization algorithm for this model by using platform, are: Shock absorption, Lead time compression and
dynamic programming. K. Inderfurth [17], S. Minner [18], supply order generation.
K. Inderfurth and S. Minner [19], S. C. Graves and S.
C. Net flow
P. Willems [20] -[21], S. Humair and S. P. Willems [22]
and T. L. Magnanti [23] all extended the work of Cruz, As for the last purpose, supply order generation is based
1991 to other production line models. M. G. Huang et al. on a net flow equation combining not only real inventory
[24] used the dynamic programming approach to solve the information, but also the supply and the demand information
BAP for a flow-shop-type production system to maximize hence the demand driven aspect. Each buffer has its own
its throughput or minimize its WIP . All these results were daily planning equation, allowing planners and managers to
established in a production environment with a safety stock get their replenishment orders right and as much as needed
logic or a simple WIP inventory along the flow line. Only for the demand. The equation is defined as:
one model was proposed for a DDMRP MTO system by
N etf low = onhand + onorder − qualif ied sales (1)
S.C. Rim et al. [25] and on which we’ve based this work
and extend it to hybrid MTO/MTS manufacturing systems • on-hand: real physical available inventory.
with better numerical experimentation. • on-order: incoming replenishment orders.
• qualified sales: client made orders and qualified demand
III. P ROBLEM STATEMENT: DDMRP B UFFER
spikes.
S TRATEGIC P OSITIONING
A. Demand Driven Material Requirements Planning D. Buffer Profiles
With the rising need of managers and decision makers Buffers in DDMRP are shaped by a three-zone scheme. A
to perform planning and execution frameworks. DDMRP three colors code signal at each decoupling point, providing
has succeeded in gathering the best practices of exiting a practical and an easy global management. The buffer zones
methods. All of which had proven their efficiency in the and levels are defined and calculated by three factors: item
field of planning and inventory management, like MRP, type, lead time and variability. The red zone stands for the
Lean and theory of constraints. But they lacked features safety stock covering the peaks of unscheduled demand or
to accommodate the high variability of demand and to get supplier’s unreliability. The yellow zone represents the main

Authorized licensed use limited to: University of Canberra. Downloaded on May 20,2021 at 23:25:11 UTC from IEEE Xplore. Restrictions apply.
demand coverage, as it is defined by the quantity of products solving method only approximate the optimal solution for
needed at hand while an order is in manufacturing process. small instances. In our proposition, we extend the model to
The green zone, seen as the replenishment orders gauge. Its minimize the costs for hybrid MTO/MTS complex manufac-
level determines the frequency, as an order is given once the turing systems and at the same time determine the profiles
net flow signal crosses from the green zone to the yellow of the enabled buffers along the bill of materials. This being
one. And since the size of each order is given to make the done by considering the Net flow equation, which includes in
signal reach the top of the green zone, we assure a minimum its formulas the expressions of the buffer profiling. Meaning
order quantity by the green zone size. This profiling actually that the dimensions of the buffer’s three zones are always
is a buffer allocation process, which comes by definition taken into consideration in the optimization process.
right after choosing the buffer position. C. Ptak, C. Smith
[1]. The DDMRP method handled this matter sequentially, G. Problem formulation
but we could do both at the same time as we’re going to see
below. Also a new formula to dimension the safety stock for The buffer positioning can be defined as a Location
these buffers was presented by C.J. Lee and S.C. Rim [27]. Inventory Problem. The process can be modeled by an
algorithm, giving the best and optimal positions along the
E. Buffer Positioning bill of materials to accommodate the demand. The prob-
To answer the question of where the decoupling point lem as handled in this paper, has an objective function
buffers should be, and as effective as possible along the of minimizing the total cost of inventory, with the buffer
supply chain, the DDMRP method has defined six key positions as a decision variable, while under the constraint
positioning factors. For the importance of this first step, the of service lead time. All of the lead times are deterministic.
factors focus on some strategic and tactical details of the The production and buffer capacities are infinite. Positioning
product qualifying it to be a decoupling point. decoupling points in different positions in a bill of materials
• Demand Lead Time: as defined by APICS’s dictionary,
makes some components make-to-stock and leave the non-
it’s ”The amount of time potential customers are willing buffered ones as make-to-order, but the aspect of the system
to wait for the delivery of a good or a service. Syn: is judged by the situation of the final product. Enabling the
Customer tolerance time. (p.45) option of having a buffer on the final product, as it may be
• The Benchmark: to assess the concurrence ability, pro-
the optimal solution in some cases, makes the model valid
viding a reduced lead time could get a better market for either MTO or MTS systems, hence the hybrid aspect.
share. Thus, a potential sales turnover growth, but For modelling issues, we assume that every component of
possibly also an increase in operational costs. the BOM is unique and is part of the assembly of one only
• Demand Forecast Horizon: how early we get the de-
parent component.
mand information, which is equal or more than the 1) Notations: The following notation is used in the paper:
demand lead time. 1) A BOM is represented by a matrix of L lines and C
• External Variability: demand and supply variability columns, where the components are located by {i ∈
• Strategic components: a key product, the most shared L, j ∈ C}.
component or simply the position allowing the best lead 2) ti,j The processing time of the component at the (i,j)
time compression. position,
• Bottleneck: components with critical manufacturing 3) ai,j The lead time of the ij th component,
process 4) parenti,j considering that every node in the BOM has
Nonetheless, these factors are checked manually and qual- its parent node in the level above (i-1), this variable
itatively for every potential decoupling point. For complex point on its column position,
bill of materials, the task would be much complicated with 5) sonsi,j,k The column position of the k th son node of
a larger margin of error. the ij th component,
6) SvT The customer tolerated service time,
F. Strategic Buffer Positioning and dimentioning 7) budget The storing costs budget,
There were efforts to automate the buffer positioning 8) adu The average daily usage of the final product,
and try to find the optimal positions using mathematical 9) rqtfi,j Required quantity of ij th component to pro-
models. The latest work being the article of J. Jiang and duce the final product,
S.C. Rim [28] with a model minimizing the storing costs 10) StC i, j Storing costs of the ij th component,
with stochastic lead times in a Make-to-Order manufacturing 11) T C The total of Storing costs for all the components,
system. This work was an extension of their previous work 12) baii,j Buffer average inventory at the node ij th ,
[25] with the same objective function for the same MTO 13) unit pricei,j Unit price of the ij th component,
system but with deterministic lead times. Although their 14) aih costi,j The average inventory holding cost rate of
model is coherent for the specific context, their suggested the ij th component,

Authorized licensed use limited to: University of Canberra. Downloaded on May 20,2021 at 23:25:11 UTC from IEEE Xplore. Restrictions apply.
15) lt f actori,j lead time factor of the ij th component, Minimize
an adaptation to the demand signal, TC =
X X
unit pricei,j · aih costi,j ·
16) var f actori,j variability factor of the ij th compo-
∀i∈L ∀j∈C
nent, for safety stock adaptation. (10)
(1, 5 + var f actori,j ) · lt f actori,j ·
The lead-time factor and variability factor are elements
introduced by the DDMRP method and integrated in the adu · rqtfi,j · ai,j · xi,j
profiling of the inventories held at each decoupling point. s.t.
The first one is usually between 20% and 40% for products a0,1 ≤ SvT
with long lead-times allowing a high frequency of order The constraint ensures that the lead-time of the final
generation. And the second one is important for products product, as a function of all the other lead-times, is less
with high variability, assuring a more important safety stock. or equal to the service time SvT . This problem is a non-
Here the decision X = {xi,j , i ∈ L, j ∈ C}, as an output linear minimization problem subject to nonlinear constraint.
of the model would be: whether a buffer is installed at each The use of DDMRP formulations makes the optimization
position or not, also meaning if the product at that station has problem of buffer positioning more complex.
an MTS or an MTO policy. We define the decision variable
H. Illustrative DDMRP positioning example
xi,j as follows:
( To demonstrate the importance of the model of strategic
0, if no buffer installed for the ij th component buffer positioning, we consider an example of a final product
xi,j =
1, otherwise FP1 and its BOM, represented by the figure 1. This product
(2) has an average daily usage of 10 units and a production lead-
The total inventory cost of the bill of materials would be: time of 11 days. Without any buffers on its components, FP1
X X is prepared in 22 days which is the cumulated lead-time of
TC = StC i, j · xi,j (3) the BOM.
∀i∈L ∀j∈C

The lead-time of a product is related to its sons and their FP1(11days)


buffer status,

th
ti,j , if the ij node has no son

P2(4days) P3(6days)
ai,j = maxk∈sonsi,j [(1 − xi+1,k )· (4)

ai+1,k + ti,j ], otherwise

For every decoupling point, in case a buffer is associated, P4(3days) P5(5days)


a profiling is done for it since it is linked to the decision
variable. The resulted three colour coded (red, yellow,green) Figure 1. The Bill Of Materials of the product FP1
levels are dimensioned by the equations :
ri,j = (1 + var f actori,j ) · lt f actori,j · adu · rqtfi,j · ai,j To guarantee a service time of 16 days, one could consider
(5) putting a buffer at the component P3, narrowing down the
yi,j = adu · rqtfi,j · ai,j (6) preparation time to 15 days. This time compression comes
with a price, the inventory cost. Giving that we only need
gi,j = lt f actori,j · adu · rqtfi,j · ai,j (7) a single component P3 to produce FP1, and at a unit price
And by definition of the method, based on the observed flow of 100 with a holding cost rate of 0.25, a lead-time factor
of replenishment orders generated by a buffer, the average of 0.5 and a variability factor of 0.25, the buffer average
onhand inventory available at each decoupled point is the inventory for it is :
sum of the red level and half of the green level. baiP 3 = (1.5 + var f actorP 3 )·
(11)
1 lt f actorP 3 · adu · aP 3 = 96
baii,j = ri,j + gi,j (8)
2 Thus the total cost would be the inventory cost for this buffer
Thus, the storing costs for a buffered component would be since it is the only one placed in the BOM.
written as:
T C = unit priceP 3 · aih costP 3 · baiP 3 = 2400 (12)
StC i, j = unit pricei,j · aih costi,j · baii,j (9)
Meanwhile the output of the model is: X={0;0;1;1;1} which
With the previous notations and equations, the mathemat- suggests to put buffers on the components P3, P4 and
ical formulation of the optimization problem can be written P5. Assuming again that we only need one piece of these
as : components. For a price of 25 for P4 and 30 for P5 and with

Authorized licensed use limited to: University of Canberra. Downloaded on May 20,2021 at 23:25:11 UTC from IEEE Xplore. Restrictions apply.
the same lead-time and variability factors as P3 at a holding And so on, we get a SAT instance of nesting clauses
cost of 0.2, we get the following buffer average inventories with N + next literals. next is the number of parts at
and costs, using the formulation in (12) and (13): the extremities of the BOM.
• Application
Table I
T OTAL COSTS OF THE OPTIMIZED BUFFER POSITIONING
Let’s consider the previous illustrative positioning ex-
ample in figure 1. The SAT formulation would be:
Component bai StC ”([(t2 ≤ Q − t1 ) OR x2 ] AN D [[(t4 ≤
P3 52 1300 Q − t1 − t3 ) OR x4 ] AN D [(t5 ≤ Q −
p4 26 130
p5 44 264 t1 − t3 ) OR x5 ]] OR x3 ) OR x1 ”
Total Cost 1694 • Resolution
Since in the positioning problem, we try different
In fact adding buffers in this case, decreased the tension combinations of buffers (xk ) to minimize the costs of
on the buffer of the component P3. For this part, which is inventories while satisfying the constraint ” a0 ≤ Q
on an upper level than P4 and P5, it is more costly to store ”, we are practically checking whether the CNF-SAT
inventory. Thanks to the buffers on P4 and P5, the decoupled above is consistent. Thus, by solving the problem we
lead-time for P3 is compressed. This results in less inventory are solving a CNF-SAT problem. Meaning the buffer
of P3, in consequence less inventory storing costs. The total positioning problem can’t be less hard to solve than the
cost with the optimized buffer positions is minimized all CNF-SAT problem.
while satisfying the constraint of the service time SvT . An
IV. H EURISTIC S OLVING M ETHOD : A MILP BASED
improvement of 29% is reported for this BOM of 5 nodes,
HEURISTIC
and it can be very considerable for larger and more complex
supply chain networks. In the proposed algorithm, first we use the linearized
model ”CostOptim” to get a lower bound of the costs.
I. Complexity Analysis Then we optimize the lead time of the final product using
Similar to the general safety stock placement problem, ”TimeOptim”, to check whether we could get feasible lead-
that was proven to be NP-hard by Lesnai E. in the article times with a cost equal to LB. Else we iterate the model
[29]. We as well proved that the problem studied in this ”TimeOptim” by a dichotomy process between LB and
paper can be considered NP-hard by a reduction of a classic m · LB (m ≥ 2).
known NP-complete problem demonstrated in the Cook-
A. Model linearization
Levin Theorem: CNF-SAT (Boolean satisfiability problem).
To describe the proposed heuristic, we consider the fol-
• Formulation of an instance of the buffer positioning
lowing linearization of the previous model. Using the same
problem:
notation, we consider the new variable :
The problem can be presented by a BOM of N parts. A
part k has a processing time tk , a decoupled lead-time zi,j = ai,j · xi,j (13)
ak and Boolean xk which is true if the buffer is enabled
or false if it is disabled. The constraint of the problem is The equation 10 would be as follows:
X X
the service time Q which is the time not be exceeded to TC = unit pricei,j · aih costi,j
produce the final product. For simplification purposes, ∀i∈L ∀j∈C
we start enumerating the parts from the top and from ·(1, 5 + var f actori,j ) · lt f actori,j · adu · rqtfi,j · zi,j
the left to the right. The final product would be k = 1. (14)
And since we also have the option to put a buffer on
the final product itself, we introduce the part 0 as a In addition to our new decision variable, several additional
parent of the final product node, which may represent constraints appear to be necessary for z. Thus the follow-
the client (t0 = 0). ing constraints would be added to our models: with M a
• Formulation of the CNF-SAT instance: sufficiently big number,
Let’s consider the Boolean expression ”a0 ≤ Q”. 1) With idxi,j indexing if the node has a son or not
This statement consistency is equivalent to the follow- ∀i ∈ L, ∀j, k ∈ C
ing:
ai,j ≥ ti,j + (ai+1,sonsi,j,k − zi+1,sonsi,j,k ) · idxi,j
”(a1 ≤ Q) OR x1 ”
(15)
Assuming every part k has Mk components, then Sk =
2)
k1...kMk the statement is equivalent to:
∀i ∈ L, ∀j ∈ C zi,j ≤ ai,j (16)
”([(t11 ≤ Q − t1 ) OR x11 ] AN D [(a12 ≤
Q − t1 ) OR x12 ] AN D...AN D [(a1Mk ≤ 3)
Q − t1 ) OR x1M1 ]) OR x1 ” ∀i ∈ L, ∀j ∈ C zi,j ≥ 0 (17)

Authorized licensed use limited to: University of Canberra. Downloaded on May 20,2021 at 23:25:11 UTC from IEEE Xplore. Restrictions apply.
4) Algorithm 1 MILP-based heuristic Algorithm
1: Function BufferPositioning(x, SvT )
∀i ∈ L, ∀j ∈ C zi,j ≥ ai,j − M · (1 − xi,j ) (18) 2: xi,j ← 0 ∀i ∈ L, ∀j ∈ C
3: budget ← CostOptim(SvT ) . A LB of the optimal
5)
cost
∀i ∈ L, ∀j ∈ C zi,j ≤ M · xi,j (19) 4: dlt ← T imeOptim(budget). The best dlt with the LB
6) as a constraint
5: T C ← ComputeT C() . The actual costs for the best
∀i ∈ L, ∀j ∈ C (1 − idxi,j ) · ai,j ≤ ti,j (20) dlt
6: m ← 1
B. Used models in the heuristic 7: if T C = budget then
With the previous notations, the mathematical formulation 8: sol ← T C
of our problem can be written as: 9: else
With 10: if dlt > Svt then
X X 11: m←2
TC = unit pricei,j · aih costi,j · baii,j · xi,j 12: while dlt > SvT do
∀i∈L ∀j∈C 13: dlt ← T imeOptim(budget ∗ m)
(21)
14: m++
baii,j = (1, 5+var f actori,j )·lt f actori,j ·adu·rqtfi,j ·ai,j 15: end while
(22) 16: sol ← ComputeT C()
• For the cost minimization model (CostOptim) 17: end if
18: n←1
M inimize TC (23) 19: do
20: if dlt ≤ SvT then
s.t. 21: k ← −1
a0,1 ≤ SvT 22: else
• For the lead-time minimization model (TimeOptim) 23: k←1
24: end if
M inimize a0,1 (24)
25: dlt ← T imeOptim(budget ∗ (m + k/2n))
s.t. 26: T C ← ComputeT C()
T C ≤ budget 27: if T C < sol and dlt ≤ SvT then
28: sol ← T C
C. The Algorithm of the heuristic 29: end if
Before describing the general algorithm the following 30: n++
procedures are needed : 31: while T C ≤ sol
32: end if
1) The procedure to compute the total cost of the en- 33: return sol . return the optimal cost
abled buffers along the BOM is using the equation 3
presented above.
2) Other CPLEX related optimization procedures are
used such as CostOptim() and TimeOptim(). A. Benchmark generation
With the previous assumption of the uniqueness of parent
V. E XPERIMENTAL STUDY nodes, we generate the problem instances as follows:
In this section, we summarize the computational exper- • We specify the number of nodes composing the BOM,

imentation with the developed algorithm using the MILP • We specify the number of levels and columns shaping

based heuristic presented previously. For comparison pur- the BOM,


poses, we solve the non linear model using the exhaustive • Randomly, we generate lead-times for the nodes,

research. Then we used MILP based heuristic. For the MILP • Following a direct material cost strategy, every node

component of the heuristic, we use CPLEX 12.8.0. An Intel has a price larger or equal to the sum of the prices of
i7-4700MQ 2.40 GHz with 8GB of RAM, with a Windows its components,
8.1 operating system is used. The choice of TypeScript as The benchmark for the set of bigger instances is on our
a programming language is to allow us to implement the perspectives as we intend to extend the BOMs to more
developed heuristic on a user friendly Web plate-form ready complex examples using an instance generator that we
to update on further perspectives. developed.

Authorized licensed use limited to: University of Canberra. Downloaded on May 20,2021 at 23:25:11 UTC from IEEE Xplore. Restrictions apply.
B. Results R EFERENCES
In this paper, we generated a set of instances by number [1] C. Ptak, C. Smith, Demand Driven Material Requirements
of nodes: small {3, 5, 6, 13}. Many shapes of BOMs Planning. Industrial Press, INC, 2016.
were generated for the small instances, adequate ones were [2] X. Y. Sun, P. Ji, L. Y. Sun, Y. L. Wang, “Positioning multiple
considered for the tests. As shown in Table II, the results decouling points in a supply network,” Int. J. Production
obtained by the developed method are similar to the optimal Economics, vol. 113, pp. 943-956, Jan. 2008.
solution found by the exhaustive research, except for the
instance of 13 components. [3] S. Weiss, J. A. Schwarz, R. Stolletz, “The Buffer Allo-
cation Problem in production lines: Formulations, solution
methods, and instances,” IISE Transactions, Feb. 2018. DOI:
Table II 10.1080/24725854.2018.1442031
C OMPARISON OF THE RESULTS

Set Nodes SvT Exhaustive research MILP heuristic


[4] G. Liberopoulos, “Comparaison of optimal buffer allocation in
TC TC flow lines under installation buffer, echelon buffer, and CON-
Small 3 6 256 256
WIP policies,” Flexible Services and Manufacturing Journal,
5 6 15600 15600 Mar. 2019. DOI: 10.1007/s10696-019-09341-y
6 6 43600 43600
13 6 66912 71950 [5] K. F. Simpson, “In-process inventories,” Operations Research,
no. 6, pp. 863-873, Nov. 1958.

The proposed heuristic gives close solutions to the exhaus- [6] S. C. Graves, S. P. Willems, “Optimizing Strategic Safety
tive method for small instances sets with more important Stock Placement in Supply Chains,” Manufacturing & Service
Operations Management, vol. 2, no. 1, pp. 68-83, Winter 2000.
computational time. However we expect that the heuristic
could be more efficient on term of computational time [7] E. Koenigsberg,“Production Lines and Internal Storage - A
for medium and big instances compared to the exhaustive Review,” Management Science, vol. 5, no. 4, pp. 410-433, Jul.
method. The iterations of the exhaustive research go through 1959.
every combination possible of the buffer positioning giving
[8] J. M. Smith, F. R. B. Cruz,“The Buffer Allocation Problem for
it an exponential complexity giving us long computational general finite buffer queueing networks,” IISE Transactions,
times. Using heuristics is a good alternative than using non vol. 37, no. 4, pp. 343-365, Feb. 2007.
linear solver, which does not always guarantee the optimal
solution. [9] P. Glasserman, S. Tayur,“Sensitivity Analysis for Base-Stock
Levels in Multiechelon Production-Inventory Systems” Man-
agement Science, vol. 41, no. 2, pp. 189-376, Feb. 1995.
VI. C ONCLUSION AND P ERSPECTIVES
[10] D. C. Whybark, S. Yang,“Positioning inventory in distribution
Through this paper, we presented an extension of the systems” Int. J. Prod. Econom., vol. 45, pp. 271-278, Aug.
strategic positioning model to hybrid MTO/MTS manufac- 1996.
turing systems. A solving method was developed through
[11] Y. Park, J. Shintaku, P. Hong, “Modularity of Flat Panel Dis-
linearization. Multiple test runs were realized on small
play TV and Operation Management Practices: A Case Study
instances. The experimental study showed that the proposed pf LG Electronics,” Manuf. Management Research Center, Jan.
heuristic gave close solutions to the exhaustive method even 2009.
with more important computational time. We expect that
exploring results on medium and large instances will show [12] M. Amiri, A. Mohtashami,“Buffer allocation in unreliable
production lines based on design of experiments, simulation,
that the heuristic will give efficient approximate solutions
and genetic algorithm,” Int. J. Advanced Manuf. Tech., vol. 62,
with reasonable time than exhaustive method of course and pp. 371-383, Sep. 2012.
the non-linear solver. This will be on our future research
direction. [13] L. Tiacci, “Simultaneous balancing and buffer allocation
decisions for the design of mixed-model assembly lines with
parallel workstations and stochastic task times,” Int. J. Prod.
ACKNOWLEDGMENT Econom., vol. 162, pp. 201-215, Apr. 2015.
This work is a part of the ELSAT 2020 project. It was [14] A. Costa, A. Alfieri, A. Matta, S. Fichera, “A parallel tabu
performed in collaboration with the ”Pôle d’excellence Lo- search for solving the primal buffer allocation problem in serial
gistique & Supply Chain en Hauts-de-France Euralogistic”. production systems,” Computers & Oper. Res., vol. 64, pp. 97-
ELSAT is co-financed by the European Union with the 112, Dec. 2015.
European Regional Development Fund, the French state and [15] S. Y. Kose, O. Kilincci, “Hybrid approach for buffer alloca-
the Council of Hauts-de-France Region. The support of tion in open serial production lines,” Comput. Operat. Res.,
ELSAT 2020 and Euralogistic is gratefully acknowledged. vol. 60, pp. 67–78, Aug. 2015.

Authorized licensed use limited to: University of Canberra. Downloaded on May 20,2021 at 23:25:11 UTC from IEEE Xplore. Restrictions apply.
[16] S. C. Graves, “Safety stocks in manufacturing systems,” 1987.

[17] K. Inderfurth, “Safety stock optimization in multi-stage in-


ventory systems,” Int. J. Prod. Econom., vol. 24, pp. 103–113,
1991.

[18] S. Minner, “Dynamic programming algorithms for multi-stage


safety stock optimization,” OR Spektrum, vol. 19, pp. 261–271,
1997.

[19] K. Inderfurth, S. Minner, “Safety stocks in multi-stage inven-


tory systems under different service measures,” Eur. J. Oper.
Res., vol. 106, pp. 57–73, 1998.

[20] S. C. Graves, S. P. Willems, “Optimizing strategic safety stock


placement in supply chains,” Manuf. Serv. Oper. Manage., vol.
2, pp. 68–83, 2000.

[21] S. C. Graves, S. P. Willems, “Erratum: optimizing strategic


safety stock placement in supply chains,” Manuf. Serv. Oper.
Manage., vol. 5, pp. 176–177, 2003.

[22] S. Humair, S. P. Willems, “Optimizing Strategic Safety Stock


Placement in Supply Chains with Clusters of Commonality,”
Oper. Res., vol. 54, pp. 725–742, Aug. 2006.

[23] T. L. Magnanti, Z.J. M. Shen, J. Shu, D. Semchi-Levi, C.P.


Teo, “Inventory placement in acyclic supply chain networks,”
Oper. Res. Lett., vol. 34, pp. 228-238, 2006.

[24] M. G. Huang, P. L. Chang, Y. C. Chou, “Buffer allocation


in flow-shop-type production systems with general arrival and
service patterns,” Computers & Oper. Res., vol. 29, no. 2, pp.
103-121, Feb. 2002.

[25] S.C. Rim, J. Jiang, C. J. Lee, “Strategic Inventory Positioning


for MTO Manufacturing Using ASR Lead Time,” Log. Oper.,
S.C.M. and Sustainability. P. Golinska, Ed., EcoProduction,
Springer, Switzerland: Basel, 2014, pp. 441-456.

[26] G. Sharman, “The rediscovery of logistics,” Harvard business


review,, vol. 62, no. 5, pp. 71-79, Jan. 1984.

[27] C.J. Lee, S.C. Rim, “A Mathematical Safety Stock Model for
DDMRP Inventory Replenishment,” Mathematical Problems in
Eng., 2019.

[28] J. Jiang, S.C. Rim, “Strategic WIP Inventory Positioning for


Make-to-Order Production with Stochastic Processing Times,”
Mathematical Problems in Eng., 2017.

[29] S.C. Graves, E. Lesnai, “Optimizing Safety Stock Place-


ment in General Network Supply Chains,” Proceedings of
Singapore-MIT Alliance Annual Symposium Conference, 2004.

Authorized licensed use limited to: University of Canberra. Downloaded on May 20,2021 at 23:25:11 UTC from IEEE Xplore. Restrictions apply.

You might also like