Sie sind auf Seite 1von 6

WINE BOTTLING SCHEDULING OPTIMIZATION

R. Berruto, C. Tortia, P. Gay

ABSTRACT. The beverage market has become very dynamic and competitive, and wine marketing strategies involve, besides
the quality aspect, the aspect of precise order delivery. Moreover, bottling plant planning is difficult, as consumer demand
is highly influenced by seasonality and many different types of packaging are requested. This article presents a new scheduling
method for planning bottling activities in modern wineries. The bottling plant has been assumed as a single-machine job shop
where wine orders of different amounts and due dates must be processed. In order to limit the complexity introduced by the
large number of variables and constraints, the sequencing has been obtained by means of a two-step procedure based on
mixed-integer linear programming (MILP) algorithms. The optimization takes into account, for each wine type, data on
production rates, storage levels, minimum batch size, labor and storage costs, and risk due to not having the minimum storage
and having lost sales. The method operates on a finite time horizon, typically four weeks, with recursive rescheduling each
week. The effectiveness of the proposed method is shown by an example using data collected in a large winery in the Piedmont
region of Italy.
Keywords. MILP, Scheduling, Wine bottling.

D
ecisions regarding bottling scheduling require mal solutions, from among which they select and refine the
knowledge of many different aspects of the win- final scheduling.
ery, such as sales, production dynamics and tim-
ing, logistics, and storage management. For these
reasons, the planning decision in big wineries is usually made
by a team. The large number of involved parameters that have
METHODS
The method proposed in this article balances computational
to be considered makes the planning difficult, so that in most
cases, a wine bottling schedule is obtained empirically, on the feasibility with the requirements of winery managers. Such a
method is based on a two-step optimization procedure. The first
basis of the personal experience of the team members.
step is performed by a mixed-integer linear programming
This article presents a new method to plan bottling
(MILP) algorithm, while the second step can be carried out
activities in wineries that minimizes costs and improves
directly by the bottling manager (fig. 1). In order to take into
overall throughput. This article considers a single-line
account the time-varying set of orders, sales manager requests,
bottling plant, where various wines are bottled in different
and bottling plant production dynamics, a time horizon of one
bottle types and packages. The scheduling of bottling
month and recursive rescheduling each week were chosen.
activities is treated as a single-machine job programming
The first step is designed to optimize the allocation of the
problem in a multi-product context.
wine orders in each of the different weeks belonging to the
The simplest way to obtain suitable scheduling consists,
assumed time horizon. This optimization is accomplished
in principle, in the direct optimization of the bottling
using the entire database, including orders, inventory, due
sequence with multiple product orders and different due
dates, minimum stocks, manpower, and processing times.
dates. However, the large number of variables and constraints
The second step produces the daily sequencing according to
involved in real applications can make this problem difficult
the output of the first algorithm, on the basis of a limited
(Mendez et al., 2000). Moreover, such an optimization will
number of orders that have been optimally selected from the
automatically displace sales managers from the production
global order database and assigned to one production week.
planning, even for small last-minute variations. Instead of
In this phase, the number of involved orders and the variables
obtaining one closed optimal solution, winery managers
to be optimized are drastically reduced, so that sequencing
seem to be most interested in obtaining a set of close-to-opti-
could be managed by the bottling manager.
This article focuses on the optimal allocation procedure
concerned in the first step, which constitutes the most
Article was submitted for review in February 2005; approved for innovative part of the proposed method. Even if the solution of
publication by the Information & Electrical Technologies Division of two subsequent optimization problems could, in principle, lead
ASABE in January 2006.
The authors are Remigio Berruto, ASABE Member, Associate to a sub-optimal solution, this method is more efficient,
Professor, Cristina Tortia, Technical Assistant, and Paolo Gay, Assistant allowing the bottling manager to have a certain degree of
Professor, Department of Agricultural, Forest and Environmental freedom in the daily scheduling while preserving the improve-
Engineering and Economy, Grugliasco, Italy. Corresponding author: ments gained by the MILP solver solution. In bottling
Remigio Berruto, Department of Agricultural, Forest and Environmental
Engineering and Economy, Via Leonardo da Vinci 44, 10095 Grugliasco
scheduling, the orders may come directly from customers, with
(TO), Italy; phone: +39-011-6708596; fax: +39-011-6708591; e-mail: associated due dates, or be automatically generated to meet
remigio.berruto@unito.it. inventory requirements for maintaining a minimum stock.

Transactions of the ASABE


Vol. 49(1): 291−295 E 2006 American Society of Agricultural and Biological Engineers ISSN 0001−2351 291
Figure 1. Layout of the proposed two-step wine bottling scheduling method.

While many research efforts have been dedicated to multiple of a batch of bottles (e.g., bottles contained in a
optimizing industrial plants scheduling (Méndez et al.,2000; carton or on a pallet). This is obtained by introducing the
Méndez and Cerdá, 2002, 2003), agro-food systems were vector b = [b1 b2 ... bnw ]T being bi , where i = 1, ..., nw, the
until now rarely considered (Banga et al.,2003). Agro-food number of bottles constituting a batch. The packed bottles are
production scheduling often imposes more difficult stored in the winery warehouse, which has a limited capacity
constraints than the manufacturing industry due to seasonal- of xmax bottles.
ity, product shelf-life, and storage conditions (Gargouri et al., Continuous time is discretized in periods whose length,
2002). typically one week, depends on production dynamics.
Limitations in resource availability must be considered Periods are numbered and labeled by an integer index. The
while planning the bottling activity, including manpower inventory of the bottles stocked at the end of the kth week is
availability during normal time and overtime, storage reported in the vector variable:
capacity of the winery warehouse, and lost sales when the
demand cannot be fulfilled. Lost sales should not exceed a x(k ) = [x1 (k ) x2 (k ) AAA xnw (k )] T (1)
certain percentage of the demand, depending on the type of
wine, or should not be allowed at all. Another constraint to while the number of batches of bottles processed during the
be considered is the wine demand and minimum stock same period k is reported in the vector:
requests, which are determined from orders, the type of wine, ξ(k )= [ξ1 (k ) ξ 2 (k ) AAA ξ nw (k )] T (2)
and historical sales data. These variables are customized, for
each type of wine, by the manager. The variable costs that The warehouse of the winery has a finite capacity. This
have to be taken into account to minimize bottling costs are can be expressed by introducing the following constraint for
related to manpower, storage, and lost sales. any period k:
In addition, for each wine type to be bottled, the following nw
information must be known: wine stock at the beginning of x (k ) 1 = ∑ x j (k )≤ xmax (3)
the first period considered in the bottling scheduling activity j =1
(data could be imported from the inventory database); At the end of the last planned period np , a minimum
bottling plant production rates for each wine type (bottling
processing times are influenced by both wine type, i.e., spar- warehouse stock of x∞ = [ x∞ 1 x∞ 2 AAA x∞ nw ]T bottles of
kling, red, or white, and type of cork, labels, and bottle wines is required. Such a bound is generally drawn up to
shape); and the number of bottles of the same wine prevent any possible loss of sales. Any lack of bottles with
constituting a batch (wine bottling is batch-oriented, and the respect to this minimum storage quantity is taken into
number of bottles present in a batch depends on wine type and account as an additional cost at two levels:
is a multiple of bottles in cartons, the quantity stored on a S Lack of bottles between 75% and 100% of the desired
pallet, or the number of bottles obtained from a whole tank quantities, which corresponds to a low risk:
of wine). In our case, the number of bottles per each batch was R75 −100 (k ) =
specified by the manager for each type of wine.
The bottling plant considered in this article has been [ R75 −100 (k ) R75 −100 (k ) AAA R75 −100 (k )]T
1 2 nw
assumed as a single machine with a daily initial setup. Setup
and failure times of different plant units have been included S Lack of bottles between 50% and 75% of the desired
in production rates. The bottling process is first described by quantities, which corresponds to a higher risk:
means of a linear discrete model that takes into account R50 − 75 (k )=
demand, stock availability, production dynamics, resources,
and costs. The optimal schedule is then obtained by solving [ R50 − 75 (k ) R50 − 75 (k ) AAA R50 − 75 (k )]T
a minimization problem using mixed-integer linear program- 1 2 nw

ming (MILP) techniques. The domain of feasibility of the Then the following constraints need to be satisfied for
solution and the constraints are extracted from the model. each kth period:
R75 −100 (k ) ≤ 0.25 x∞ (4)
MODEL FORMULATION
The winery processes nw different commercial products.
For handling purposes, the lot size can be defined as a R50 − 75 (k ) ≤ 0.25 x∞ (5)

292 TRANSACTIONS OF THE ASABE


and the following inequalities must be respected: The result consists in the sequence of vectors (k), k = 1, ... ,
np .
x(k )+ R50 − 75 (k )+ R75 −100 (k )≥ x∞ (6) The minimization problem can be expressed as:
The wine demand, expressed as batches of bottles to be min z (ξ )
delivered at the end of the kth period, is indicated as (k) = ξ(k ), k =1,2,K , n p
(13)
[1(k) 2(k) ... nw (k)] T. Analogously, the vector indicating subject to
the resulting lost sales for each period k, in terms of bottles, eqs. 3, 4,5, 6,8,
9 ,10,11,12
is defined by u(k) = [u 1(k) u 2(k) ... u nw (k)]T. Therefore, the
inventory at the end of the kth period is given by: where z( ), reported below, is the cost function to be mini-
mized:
x(k ) = x (k − 1)+ b o ξ(k )+ u(k )− v (k ) (7)
np np
for any k > 0, where x(0) is the wine stock at the beginning z = ∑ L ord (k )c ord + ∑ L r (k )c r
k =1 k =1
of the scheduled window, and b o ξ(k )=
[b1ξ1(k ) b2ξ2 (k ) AAA bnwξ nw (k )] indicates the Hada-
T np np
+ ∑ c Tx x (k ) + ∑ c uT u (k )
mard product of b and (k) (Bronshtein et al. 2003). k =1 k =1
Since the inventory x(k) is non-negative for any k, the np
following sales-demand constraint can be formulated: + w 75 − 100 ∑ c uT R 75 − 100 (k )
k =1
x(k − 1)+ b o ξ(k )+ u (k )≥ v (k ) (8)
np
The bottling of each wine requires a specific processing + w 50 − 75 ∑ c uT R 50 − 75 (k ) (14)
time. Times required for a unit bottling for each wine are k =1

stored in the vector Tb = [Tb Tb AAA Tb nw ] . Since the


T
1 2 Such a cost function embeds single cost factors that have
model implements a fixed setup time (Ts ) for each period k, been introduced in the previous section.
the labor balance results in: The cost function (z) as well as the constraints (eqs. 3, 4,
5, 6, 8, 9, 10, 11, and 12) are linear in the (integer) parameters
Lord (k )+ Lr (k )= (b o Tb )o ξ(k )+ Ts (9) (k), k = 1, ..., np . Therefore, the minimization problem in
where Lord (k) is the labor request during period k in the ordi- equation 13 can be solved using any mixed-integer linear
nary labor time, and Lr (k) is the overtime labor request in the programming (MILP) software based on the branch and
same period. bound algorithm.
The following constraints limit the labor in each period k:
EXAMPLE: MODEL TUNING AND BOTTLING OPTIMIZATION
Lord (k )≤ Lord (k ) (10) IN A WINERY
This section reports an application of the proposed method
Lr (k )≤ Lr (k ) (11) to the bottling planning of a winery. All the input data, like
number and type of wines, production times, orders, and
where Lord (k ) and Lr (k ) are the maximum time values avail- warehouse capacity, were collected for a large winery in the
able in normal time and overtime, respectively, in each peri- Piedmont region of Italy. As already mentioned, for technologi-
od k. cal and marketing purposes, wines are bottled using different
Finally, the lost sales should not exceed, in any period k, type of bottles (in terms of shape, capacity, color), labels, corks,
a given umax = [umax1 umax 2 AAA umax nw ]T percentage of and other accessories. In this application, more than 200
different commercial products have been divided into homoge-
the demand, that is: neous groups in terms of wine type and bottle type, neglecting
u (k )≤ umax o v (k ) (12)
different labeling, corks, and other accessories. This grouping
results in 58 classes of products (i.e., nw = 58), hereafter referred
to as “formats.” In order to have real data for the labor
OPTIMIZATION requirement for each wine, the average bottling processing time
The storage of each wine implies a cost for each format was recorded from observation. The stock of
cx = [ c x cx AAA c x ]T , which is expressed in Euros each format at the beginning of the considered time horizon was
1 2 nw
per bottle per period. The cost, in Euros per bottle, due to lost taken from the database of the warehouse, while demand data
were taken from customer orders and historical data. All the
sales is stored in the vector cu = [cu 1 cu 2 AAA cu nw ] ,
T
other data were provided by the manager of the bottling plant.
while manpower cost during normal time (cord ) and manpow- The optimization (eq. 13) was performed to schedule the
er cost during overtime (cr ) are expressed in Euros per minute bottling activities over np = 4 periods. Each period in this
for the entire bottling team wage. The additional cost that has study represent the activity of one week. The parameters for
to be introduced whenever the minimum wine stock is not the MILP model are presented in table 1. Each row of the
satisfied can be computed as a percentage (w75−100 and table contains information about a single wine format. In this
w50−75, with w75−100 < w50-75) of the lost sales cost. example, kosher wines (e.g., formats 4 and 9) have an
The optimal scheduling is obtained solving recursively, associated minimum final stock equal to zero because they
for each period k, a np step-ahead finite horizon optimization. are delivered as soon as bottled.

Vol. 49(1): 291−295 293


Table 1. Input parameters for the model. For simplicity, only 16 formats of the 58 considered
are reported, but the optimization was performed considering nw = 58 different products.
Bottling Wine Demand (ν) Storage Initial Minimum
Processing (bottles) Cost Stock Final Stock
Time (Tb ) (cnp ) (x0) (x8)
Format (min/bottle) Week 1 Week 2 Week 3 Week 4 (Euros/bottle) (bottles) (bottles)
1 Asti spumante 0.75 0.0146 180 150 0 120 0.003 8000 6200
2 Asti 1.5 0.0492 0 0 0 14000 0.003 4990 5000
3 Asti 0.75 - 12 0.0150 0 0 0 45000 0.003 28800 29000
4 Asti kosher 0.0155 12000 0 0 0 0 0 0
5 Asti 3.75 0.0109 0 0 0 15000 0.003 2200 2200
6 Barbera 0.75 0.0144 60 60 60 2460 0.002 23490 5800
7 Barbera Asti 0.0119 1920 120 420 4080 0.002 34780 12800
8 Barbera Ceppi Storici 0.0222 300 300 0 0 0.002 9380 5400
9 Barbera kosher 0.0107 0 8000 0 0 0 0 0
10 Barbera Croia 0.0127 10000 0 0 0 0.002 22500 20000
11 Barbera Albera 0.0186 0 0 2000 10000 0.002 4920 9600
12 Piemonte Barbera 0.0290 1680 0 0 4000 0.002 41850 10000
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
55 Piemonte Cortese 0.0213 10000 20000 50000 30000 0.002 6000 0
56 Piemonte Moscato 0.0170 0 0 10000 0 0.002 5300 0
57 Dolcetto Monferrato 0.0135 0 0 0 0 0.002 3900 2400
58 Moscato Bordolese 0.0124 0 0 0 0 0.003 40700 0

The optimal schedule was computed and is presented in wine, the manager can force the scheduling algorithm to
the table 2, where the number of bottles to be filled for each fulfill the demand by increasing the cu coefficient.
week is shown. The first stage of the proposed method does The concept of expressing the demand as minimum stock
not detail the bottling sequence within each day of the week, adds flexibility to the production planning. This is very
since that relies on the manager’s decisions. As can be seen important in the case of seasonal products like Christmas
in table 2, the application of the method has generated some wines. In our case study, even if all the manpower was used,
lost sales, related to one inexpensive wine bottled (Piemonte as well as some of the overtime labor available (not shown in
Cortese). Because the objective was to minimize costs, the the table), the demand was fulfilled using part of the
manager has, in this case, an economic benefit from lost minimum stock.
sales. As can be seen in some cases (formats 2 and 3), the The schedule of the following weeks can be obtained by
MILP model provides a solution (table 2) where the recursively solving the optimization (eq. 13) at the beginning
minimum final stock is below the requested level (table 1) of each week, over a four-week period, in order to take into
because this provides the minimum bottling cost for the account new orders, to update stock availability, and to refine
winery. If lost sales are not convenient for a specific type of the demand forecast.
Table 2. Results of the optimization: bottling planned activities per week, lost sales, stock deficit, final stock per
wine at the end of the period. As in table 1, only 16 of the 58 formats are reported for ease of understanding.
Wine Bottling Scheduling (ξ)
Lost Sales Stock Deficit Final Stock
(bottles)
(unp ) (R50−75+ R75−100)np (xnp )
Format Week 1 Week 2 Week 3 Week 4 (bottles) (bottles) (bottles)
1 Asti spumante 0.75 0 0 0 0 0 0 7550
2 Asti 1.5 0 0 0 11880 0 2130 2870
3 Asti 0.75 - 12 0 1320 0 36960 0 6920 22080
4 Asti kosher 12210 0 0 0 0 0 210
5 Asti 3.75 0 0 0 14520 0 480 1720
6 Barbera 0.75 0 0 0 0 0 0 20850
7 Barbera Asti 0 0 0 0 0 0 28240
8 Barbera Ceppi Storici 0 0 0 0 0 0 8780
9 Barbera kosher 330 7920 0 0 0 0 250
10 Barbera Croia 7260 0 0 0 0 240 19760
11 Barbera Albera 4620 0 2640 7260 0 2160 7440
12 Piemonte Barbera 0 0 0 0 0 0 36170
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅
55 Piemonte Cortese 5280 18480 51480 29040 240 0 520
56 Piemonte Moscato 660 0 4620 0 0 0 580
57 Dolcetto Monferrato 0 0 0 0 0 0 3900
58 Moscato Bordolese 0 0 0 0 0 0 40700

294 TRANSACTIONS OF THE ASABE


CONCLUSIONS S Development of a second model that optimizes the se-
This article has presented a two-step method, including quence of bottling within the same week, taking into
MILP optimization, for planning bottling activities. The account the setup times that occur while switching be-
scheduling methodology takes into account the practical tween different wines and formats.
aspects of winery production planning, leaving a certain
degree of freedom to the bottling manager. ACKNOWLEDGEMENTS
Since the results of the optimization are sensitive to the We thank Araldica winery, Castelboglione (Asti, Italy),
input data, this tool can help managers improve their own for their kind cooperation to this study.
knowledge about the effect of the different parameters on the
final costs and bottling schedule. The proposed methodology
can also be used to compare alternatives for bottling planning REFERENCES
in terms of work organization, manpower allocation, and Banga, J. R., E. Balsa-Canto, C. G. Moles, and A. A. Alonso. 2003.
warehouse management policies. The method has been Improving food processing using modern optimization methods.
tested in a large winery in the Piedmont region of Italy. The Trends in Food Science and Tech. 14(4): 131-144.
optimization output resulted in a feasible schedule for Bronshtein, I. N., K. A. Semendyayev, G. Musiol, and H. Muehlig.
bottling activities. 2003. Handbook of Mathematics. 4th ed. Berlin, Germany:
In bottling scheduling, the operational time horizon, with Springer-Verlag.
the ability to update the schedule every week, as well as the Gargouri, E., S. Hammadi, and P. Borne. 2002. A study of
scheduling problem in agro-food manufacturing systems.
ability to account for expected demand as minimum stored
Mathematics and Computers in Simulation 60(3-5): 277-291.
quantity, resulted in a suitable approach. This allows the Méndez, C. A., and J. Cerdá. 2002. An efficient MILP
production limits to be circumvented during high-demand continuous-time formulation for short-term scheduling of
periods if the demand is seasonal, because even if specific multiproduct continuous facilities. Computers and Chemical
orders are not already available, the schedule plan can be Eng. 26(4-5): 687-695.
adjusted. Méndez, C. A., and J. Cerdá. 2003. Dynamic scheduling in
Further improvements will focus on: multiproduct batch plants. Computers and Chemical Eng.
S Implementation of a model for management resource 27(8-9): 1247-1259.
planning of the raw materials (corks, labels, bottles, Méndez, C. A., G. P. Henning, and J. Cerdá. 2000. Optimal
wine) used by the bottling process, which need to be scheduling of batch plants satisfying multiple product orders
with different due dates. Computers and Chemical Eng.
purchased in a timely manner.
24(9-10): 2223-2245.

Vol. 49(1): 291−295 295


14 TRANSACTIONS OF THE ASABE

Das könnte Ihnen auch gefallen