Professional Documents
Culture Documents
1 s2.0 S1474667016308825 Main
1 s2.0 S1474667016308825 Main
A new formulation for a variant of the Capacitated Lot Sizing and Scheduling
Problem with Sequence Dependent setup costs and times: Single machine, multiple
products and periods
I. Chaieb Memmi*, S. Hammami*
* ESTI, 45 Rue des Entrepreneurs, Charguia II, 2035 Tunis Carthage, Tunisia
(Tel: 216-71-941-579; e-mail: memmi_i@yahoo.fr, sondeshammami@yahoo.fr )
Abstract: We present in this paper a new formulation of the Capacitated Lot sizing and Scheduling
Problem with Sequence Dependent set up cost and times (CLSPSD) problem, well known as an NP-Hard
problem. We formulate the problem as a non linear model and use it to modify a lot sizing problem
described in literature and applied to a real-life problem of a Canadian paper maker where the paper
machine represents the bottleneck resource and consequently requires an efficient production planning
approach optimizing resource utilization and minimizing production, setup and inventory costs. In these
previous works, the dependency of the costs and times with the production sequences is not thoroughly
dealt with. We prove that by really treating the problem of determining the production sequences, we
handle to considerably reduce the supply chain total cost. This cost includes production costs, inventory
holding costs, start up and setup costs. The reduction in the total cost varies from 6 to 52% when setup
costs vary from 20% to 200% of inventory holding costs.
Let us define in what follows the variants of lot sizing sequences. From this set and in each period, the best
problems mentioned in Table 1. sequence is retained regard to its cost and whether or not each
item’s inventory is empty. Well, even if the tailor-made
The single item uncapacitated lot sizing problem (ULS): In enumeration method of the branch-and-bound type solves
its simplest form, the dynamic lot sizing problem tries to find problem instances optimally, the important number of
a production plan that minimizes the total costs of assumptions hindrances applying this method in more general
production, set up and inventory for one item over a multi- cases. In fact, the authors assume that:
period horizon. This plan has to satisfy deterministic demand.
- if the production of an item starts in a period then
The capacitated multi-item lot sizing problem (CLSP) is a the inventory of the item must be empty at the end
typical example of a large bucket model, where many of the previous period,
different items can be produced on the same machine in one
time period. The production capacity of this machine is - the set up is preserved between two adjacent
limited. periods, this means that we must first produce the
item i at period t if this item is the last one produced
Small bucket models differ from the large bucket ones by an in period (t-1),
assumption made, which imposes that, during each time
period; at most one type of item can be produced on the same - for each item, at most one lot will be produced in a
machine. This is the case for the Continuous Set Up Lot period.
Sizing Problem (CSLP).
Due to this last assumption, we could no more optimize the
The Proportional Lot Sizing and Scheduling Problem (PLSP) quantities to produce from each item. In fact, optimizing
relaxes the restriction of allowing production for only one these quantities may imply to produce the quantities
product in each time period. Nevertheless, in the PLSP, at demanded over two periods, in only the first period.
most two different items can be produced in each time period.
There is still at most one set up in each period, but the set up In (Kovács et al., 2009), the authors develop an approach
from the previous period can be carried over to the next partly based on the work of (Haase and Kimms, 2000). They
period. The Discrete Lot Sizing and Scheduling Problem address the capacitated lot-sizing and scheduling problem
(DLSP) differs from the CSLP in that a discrete production with sequence-dependent setup times and costs (CLSPSD).
policy is assumed, implying that an item must be produced at They show that the complexity of this large-bucket lot-sizing
full capacity. problem originates from the series of implicit sequencing
problems that have to be solved for the items produced in
In a Multi-level Lot Sizing problem, the production planning each time period. Consequently, they develop an efficient
is not only done at the final level for the finished product, but algorithm in order to determine during pre-processing all
also for the components and subassemblies needed to make item sequences that could appear in an optimal solution. A
this finished product. Production at one level leads to demand novel MIP formulation of CLSPSD that relies on a compact
for components at a lower level, and at the highest level, representation of those sequences by using item-related
production is triggered by market demand. binary variables is introduced. According to authors, the
model developed allows solving in reasonable time instances
In addition to this set of models described in (Jans and where the product of the number of items and number of time
Degraeve, 2007), authors in (Salomon et al., 1997), consider periods is at most 60–70.
another type of lot sizing problems, called the Discrete Lot
sizing and Scheduling Problem with Sequence Dependent set As in (Haase and Kimms, 2000), and both using MIP
up cost and times (DLSPSD). So, as it is underlined in the techniques, exact optimisation approaches to CLSPSD have
designation of this lot sizing problem type, setup costs and been suggested by (Gupta and Magnusson, 2005). The
times depend on the production sequences retained. To solve authors define a MIP that simultaneously makes the lot-
this problem, and based on a transformation of the DLSPSD sizing, and all sequencing decisions within time periods. The
into a Travelling Salesman Problem with Time Windows, an developed approach allows finding optimal solutions only for
exact solution procedure is developed. The sensitivity of the very small instances (three items and three time periods). For
proposed approach’s performance is analysed for different larger problems, a heuristic procedure is proposed in the
problem characteristics: problem dimension, inventory same paper. The authors recommend using this procedure
holding costs, set-up times and production capacity when there are many more products than there are planning
utilization. Even if the set-up times are not explicitly taken periods. From a practical point of view, the procedure
into account in the definition of production capacity comforts firms which need to solve the problem repeatedly
utilization, the authors have investigated the relationship for a few periods at a time.
between this capacity and the average CPU-time of their
algorithm. In (Bouchriha et al., 2007), a real case study is presented. The
authors discuss a specific lot sizing problem and different
The authors, in (Haase and Kimms, 2000), deal with lot scenarios where a predetermined production sequence has to
sizing and scheduling for a single-stage, single-machine be maintained, all types of products have to be produced in
production systems where set up costs and times are sequence each production period, etc. The production is planned under
dependent. They formulate a large-bucket mixed integer a cyclic manner. For this, a new approach is proposed to first
programming which is based on the set of efficient compute the cycle time by assuming constant demand, then,
84
MCPL 2010
Coimbra, Portugal, Sept 8-10, 2010
lot sizes are determined for each product within each cycle in I i0 = K i = 1,..., n (1.4)
order to satisfy demand. When comparing the different
scenarios analysed: fixing a cycle for all products, or the I it ≥ SS i = K i = 1,..., n t = 1,..., T (1.5)
quantity to produce in each cycle, etc., the authors show that ρ it ∈ {0,1} i = 1,..., n t = 1,..., T (1.6)
the scenario fixing each of the cycle and the quantities is the
most expansive. The other scenarios aren’t considerably Qit ≥ 0, I it ≥ 0 i = 1,..., n t = 1,..., T (1.7)
different in terms of potential gains. Among theses scenarios,
the last one is the less constrained, in fact, all the products The constraints include material balance (1.1), production
types have not to be produced in each period, besides, the output capacity of machine (1.2) and constraints requiring
production sequences have not to be prefixed. Even though setup in periods in which production occurs, constraints (1.3)
these assumptions are removed, this scenario does not and (1.6). Constraint (1.4) implies that an initial inventory
rigorously illustrate the CLSPSD problem, we’ll explain must be fixed in the planning problem. The authors define
more in detail this point of view when describing the this parameter according to the average of the demand.
mathematical model developed by the authors. Constraints (1.5) imply that, at the end of each period, the
stock level must be at least equal to the safety stock level.
We aim in this paper to continue the works presented in Backorders are not allowed, (1.7).
(Bouchriha et al., 2007) and specifically the last scenario
described. We then modify the lot sizing problem on a Here, we need to underline that constraints (1.3) impose that
Capacitated Lot sizing and Scheduling Problem with a minimum quantity has to be produced if a setup occurs.
Sequence Dependent set up cost and times, and consider the From this Model, we’ll omit in the subsequent paragraph the
single machine variant of this problem, multiple products, terms Kit.ρit, from the objective function, keep constraints
multiple periods and dynamic demands. Our objective is to (1.1), (1.2), (1.4), (1.5), (1.7) and modify constraints (1.3)
really focus on set up costs and times depending on and (1.6). Our objective is to introduce setup costs incurred
production sequences. The model we develop determines in each time we switch from one item’s type to the next.
each production period, the lot sizes and the production
sequences. These sequences have not to be respected from Table 2. Model (1) Parameters
one period to another. Moreover, we are not obliged to
T Period index (t∈{1, …, T})
produce all types of products in each period. The set of items
to produce in each period will be dependent on the IP Set of products to be manufactured on the machine
optimization of the production batches and different costs. (IP⊂{1, …, n})
i Product index, i∈IP
To recap, we simultaneously try to find, at each period of the hit Inventory holding cost for product i ($/ton per
planning horizon, the finished products quantities to period), i∈IP
manufacture and the production sequences to adopt. Finding bit Unit production cost ($/ton), i∈IP
these production sequences lead us to deal simultaneously Kit Setup cost for product I ($), i∈IP
with two planning levels: tactical and operational.
M A very large number
qit Minimum quantity to produce when a setup occurs
3. MATHEMATICAL MODEL
(ton)
dit Demand of item i that corresponds to period t
3.1 Initial Mathematical model (ton), i∈IP
kit Setup time for item i at period t (hour), i∈IP
The model described in (Bouchriha et al., 2007) and noted ait Unit of capacity consumption of item i at period t
Model (1), determines optimum production batch size and (hour/ton), i∈IP
setup for each planning period that minimizes the total Ct Production capacity per period (hour machine)
inventory, setup and production costs for all products over
SSi Safety stock for product i (ton), i∈IP
the entire planning period. Model (1) is a mixed integer
Ii0 Stock level at the end of the first period (ton),
program described, using notations in Tables 2 and 3, as
follows: i∈IP, It is a constant (K)
Model (1):
Table 3. Model (1) Decision variables
Min ∑ ∑ (hit . I it + bit . Qit + K it . ρ it )
T n
(1) ρit Binary variables that corresponds to 1 if the
t =1 i =1
product is manufactured at period t and 0
Subject to:
otherwise, i∈IP
Qit + I it −1 = I it + d it i = 1,..., n, t = 1,..., T (1.1) Qit The quantity of product i to produce ate period t
n
(ton), i∈IP
∑ ( k it ρ it + ait Qit ) = C t t = 1,..., T (1.2) Iit Stock level of product I at the end of the period t
i =1
(ton), i∈IP
qit . ρ it ≤ Qit ≤ M . ρ it i = 1,..., n, t = 1,..., T (1.3)
85
MCPL 2010
Coimbra, Portugal, Sept 8-10, 2010
3.2 Mathematical model for the setup period (t+1) verifies ∑ xl i t +1 = 0 what means only one
i
Taking into account setup costs within production periods product is manufactured after product l, at the period (t+1).
and between period t and period (t+1) involves combining the
following Model to the Model (1) described previously. The Table 4. Model (2) Parameters
notations used here are explained in Tables 4 and 5.
csi Start up costs of product i, i∈IP
⎛n n ⎞ T −1 n n
Min∑ ⎜ ∑ ∑ cijt . xijt ⎟ + ∑ ⎛⎜ ∑ ∑ cklt . y klt ⎞⎟
T cij Setup costs incurred when manufacturing item j
(2)
t =1 ⎝ i =1 j =1, j ≠i ⎠ t =1 ⎝ k =1 l =1,k ≠l ⎠ after item i ($), i, j∈IP
ckl Has the same signification that cij, it is used here
Subject to: just to underline the fact that this setup cost is
incurred between two distinguished periods ($), k,
zit ≤ Qit ∀i ∀t (2.1)
l∈IP
Qit ≤ zit .M ∀i ∀t (2.2)
0 ≤ Posit ≤ ∑ zit ∀i ∀t (2.3)
Table 5. Model (2) Decision variables
i
Posit ≤ zit .M ∀i ∀t (2.4) xijt Binary variables that correspond to 1 if the product
j is manufactured immediately after product i at
zit ≤ Posit .M ∀i ∀t (2.5)
period t and 0 otherwise, i, j∈IP, t∈{1, …, T}
zit + z jt ≤ 1 + Pos jt − Posit ∀i ∀j ≠ i ∀t (2.6) yklt Binary variables that correspond to 1 if product k
is the last product of the production sequence in
1 − xijt ≤ z jt − zit + ( Pos jt − Posit ) − 1 ∀i ∀j ≠ i ∀t (2.7) period t and l the first product of the production
sequence in period t+1, and 0 otherwise, k, l∈IP,
( z kt + zl t +1 + (1 − ∑ xkit ) + (1 − ∑ xil t +1) − 3) ≤ y klt .M (t∈{1, …, T})
i i
(2.8)
∀k , l ∈ {1,.., n}, l ≠ k , ∀t = 1,..., T − 1
zit Binary variables that correspond to 1 if product i is
manufactured in period t and 0 otherwise, i∈IP,
Qit ≥ 0, I it ≥ 0 ∀i, ∀t (2.9) (t∈{1, …, T})
z it ∈ {0,1},
Posit Integer variables that indicate the position of
Posit ∈ IN , ∀i ∀t (2.10)
product i in the production sequence, at period t,
xijt ∈ {0,1}, ∀i ∀j ≠ i ∀t (2.11) i∈IP, (t∈{1, …, T})
si Binary variable that correspond to 1 if product i is
y klt ∈ {0,1}, ∀k ∀l ≠ k ∀t = 1,..., T − 1 (2.12)
the first product manufactured in period 1 and 0
Constraints (2.1) and (2.2) indicate that if we manufacture otherwise, i∈IP
product i in period t then variables zit equal 1, thus, if
amounts Qit are strictly positive then zit equal 1. Relation
between variables zit and Posit are determined by constraints As we can see in Model (2), due to the absolute values, the
(2.3), (2.4) and (2.5). While constraints (2.4) and (2.5) affect mathematical model is a mixed integer non linear program.
0 to Posit when product i isn’t manufactured at period t, i.e. Solving this model with a commercial solver like LINGO
when zit equal 0, constraint (2.3) determine their domains of may give local optimum solutions.
variation. These domains are delimited by 0 and the number
of products manufactured at period t, i.e. the sum of zit. 2.3 CLSPSD mathematical model
Constraints (2.6) avoid attributing the same position in the
production sequence to two different items. In fact, if we The CLSPSD problem is finally formulated as follows:
manufacture products i and j at period t, i.e. if zit and zjt equal
1, then (Posjt-Posit) is necessary greater than 1 or lower than Model (3):
(-1). Constraints (2.7) attribute the value 1 to variables xijt
when products i and j are manufactured at period t (zit and zjt ⎛n ⎞
⎜ ∑ csi . si + ⎟
equal 1) and item’s j position succeeds item’s i position in the ⎜ i=1 ⎟
production sequence, what means product j is manufactured ⎜T ⎛n ⎟
⎞
Min⎜ ∑ ⎜ ∑ (hit . I it + bit . Qit ) + ∑ ∑ cijt . xijt ⎟ + ⎟
n n
immediately after product i (Posjt-Posit=1). Finally constraints (3)
(2.8) imply that if product k has the last position in the ⎜ t =1 ⎝ i=1 i =1 j =1, j ≠i ⎠ ⎟
production sequence at period t and product l has the first ⎜ T −1⎛ n n ⎞ ⎟
position in the production sequence at period (t+1) then the ⎜ + ∑ ⎜ ∑ ∑ cklt . y klt ⎟ ⎟
binary variable yklt equals 1. The last product of the ⎝ t =1 ⎝ k =1 l =1,k ≠l ⎠ ⎠
production sequence at period t verifies ∑ xkit = 0 what Subject to:
i
means no product is manufactured after product k, at the (1.1), (1.2), (1.4), (1.5), and (1.7), (2.1) to (2.12). We have to
period t. The first product of the production sequence at note that in constraints (1.2), variables ρit are substituted by
zit.
86
MCPL 2010
Coimbra, Portugal, Sept 8-10, 2010
4. RESULTS Since the holding costs are unitary, we have estimated total
holding costs by multiplying the random variables chosen
In this section, we focus on two points; first, we compare our from the interval [1,10] (Tale 6) by the average demand.
results with those published in (Bouchriha et al., 2007), Choosing the average demand is due to the fact that in one
secondly, we analyse the sensitivity of these results to two hand, we have no idea about the average level of stocks; in
parameters: inventory costs and setup costs. the other hand the average demand can comfort the two
opposite scenarios: important unitary holding costs, low
4.1 Comparison with the initial model (1) unitary holding costs. In fact, when the inventory holding
costs aren’t important, we have tendency to increase the
In this paragraph and in order to compare our results to those amounts stocked; in opposite, if the holding costs are
described in (Bouchriha et al., 2007), we use the illustrative important we try to reduce as possible the amounts stocked.
example given Table 6. The obtained results are given Table 7. If we observe
Scenario 2, for Model 1, the way of thinking is, since the
Table 6. The illustrative example holding costs are not very important, we have to produce, if
the machine’s capacity allows, all the needed amounts of
T 2
products and if it is possible, one product type per period (in
IP 3 our example, we have only two periods, this is why P1 and
dit Random number that belongs to the interval P2 are produced on the same period). Well, by solving
[10,200] simultaneously the lot sizing problem and the scheduling
Ii0 Random number that belongs to the interval [0,40] problem, we obtain different production sequences that lead
SSi Random number that belongs to the interval to a decrease in the total cost of the supply chain.
[0,100]
Ct 100 000 hours It is the same for scenario 1 where setup costs are not very
bit Random number that belongs to the interval [1,10] important, and so we can produce, at each period all the
hit Random number that belongs to the interval [1,10] products’ types; this is well illustrated in Model 1. In fact, we
cij and Random number that belongs to intervals defined produce two types of products in period 1 and three in period
ckl proportionally to the intervals of hit. 2. Our Model proves that simultaneously solving the lot
Kit Random number that belongs to intervals defined sizing and scheduling problems don’t systematically lead to
proportionally to the intervals of hit. the Model’s 1 conclusion. In fact, by producing three types of
qit Random number that belongs to the interval products on period 1 and only one on period 2 we can divide
[50,150] by two the supply chain total cost.
ait Random number that belongs to the interval The same reasoning can be applied to study production costs
[10,20] versus setup costs.
kit Random number that belongs to the interval [1,5]
M 109
Here, we analyse two scenarios: important holding costs and
low holding costs. Before, we should explain our interest to
87
MCPL 2010
Coimbra, Portugal, Sept 8-10, 2010
Table 7. Results of the illustrative example 4.2 Analysis of the model’s sensitivity
Model 1 Model 4 In this paragraph, we want to compare our results with those
Scenario
1 of Model 1 when varying the setup costs from 20% to 200%
1st period (P1,P2) P3-P2-P1/ of the holding costs. Fig.1 shows that, in all cases, we
- Qit (105 P1,100 P2) (1 P3,265 P2,205 P1) manage to significantly reduce the total cost. This reduction
- Iit (5 P1,40 P2,10 P3) (105 P1,205 P2,11 P3) varies from 6 to 52%. Reducing these costs can be better
2nd period (P1,P2,P3) P3 illustrated when we consider more production periods. As in
- Qit (100 P1,165 P2,150 P3) 24 P3 (Haase and Kimms, 2000) and from a practical point of view,
- Iit (5 P1,5 P2,130 P3) (5 P1,5 P2,5 P3) our model comforts more firms aiming to solve the CLSPSD
problem repeatedly for few periods at a time.
Total cost 5102 $ 2432 $
12000
Scenario
2 10000
1st period (P1,P2) P2-P1/
8000
- Qit (205 P1,265 P2) (65 P2,105 P1)
- Iit (105 P1,205 P2,10 P3) (5 P1,5 P2,10 P3) 6000
nd 4000
2 period (P3) P1-P2-P3
- Qit (105 P3) (100 P1,200 P2,25 P3) 2000
- Iit (5 P1,5 P2,130 P3) (5 P1,5 P2,5 P3)
0
20%
40%
60%
80%
Total cost 8568 $ 5653 $
100%
120%
140%
160%
180%
200%
Total cost for CLSPLD Model
Total Cost for Model I
88