You are on page 1of 18

EUROPEAN

JOURNAL
OF OPERATIONAL
RESEARCH
ELSEVIER European Journal of Operational Research 81 (1995) 17-34

Case Study

A Decision Support System (DSS) for agricultural pesticide


production planning
A.A. Elimam
Business Analysis and Computing Systems, Collegeof Business, San Francisco State University, San Francisco, CA, USA
Received August 1993; accepted April 1994

Abstract

The paper presents a Decision Support System (DSS) for the production planning of Agricultural pesticide. The
unique features of pesticide production are identified and the resulting planning trade-off's are explained. The DSS
consists of six loosely coupled modules. A mathematical programming model forms the basis of the DSS, whose
objective is to minimize the total cost of production. The model is also designed to capture seasonal demand
fluctuations, variations in labor cost, source, cost and availability of raw material and plant operating mode. The
DSS is illustrated by planning the manufacturing of nine products on two production lines to satisfy demand over
eighteen months. The system has been in use for pesticide production planning in a major US plant for more than
five years. Such use helped in rationalizing production planning decisions and led to tangible savings in production
and inventory cost, over the duration of its use.

Keywords: Seasonal demand; Labor cost; Capacity; Piecewise linear cost functions; Operating modes

1. I n t r o d u c t i o n

The objective of this p a p e r is to present a Decision Support System (DSS) for the production planning
of agricultural pesticide. The DSS is based on a mathematical programming model whose criterion of
optimization aims at minimizing the total cost of production.
A major Petrochemical Corporation produces several formulations of agricultural pesticides. D e m a n d
for pesticide follows a seasonal trend based on the needs for agricultural crop protection. D e m a n d is
usually given in specified sizes of each formulation. The plant relies heavily on union labor for its
workforce. Labor wages are adjusted once a year based on the outcome of the company-union
negotiations. R a w material availability vary according to the source of supply. The plant can select
among two operating modes 'regular or overtime'. The plant per unit cost decreases with the increase in
production quantities.
T h e issue of production planning and control in industrial facilities has b e e n treated extensively in the
literature. Recently the use of advanced computing tools such as decision support systems has been
highly emphasized.

Elsevier Science B.V.


SSDI 0377-2217(94)00090-Y
18 A.A. Elimam / European Journal of Operational Research 81 (1995) 17-34

Mattsson et al. (1989) presented a comparative study among companies on the computerization of
Manufacturing Planning and Control (MPC) systems. The study showed the status of minicomputer-based
systems in 59 companies in Sweden. In a survey article, Ovrin, Olhager and Rapp (1990) studied the use
of computer-based manufacturing planning and control (MPC) systems in 12 small swedish companies.
Their work explored: (1) improvements in the efficiency of the companys' operation, (2) reasons for
purchasing or modifying an MPC system, (3) educational efforts involved, and (4) personnel attitude and
behavior towards the MPC system. The application of MPC system in one company has been discussed in
detail. Sridharan et al. (1989) presented another survey paper to explore the impact of interaction
between the manufacturing environment and the planning and control system on plant performance. The
authors assessed manufacturing performance using measures of inventory turnover, customer service
level, and delivery lead time. Manufacturing firms have been classified based on the production process,
production volumes, and the type of the planning and control system used in a sample of companies
surveyed by mail. A random coefficient regression model was developed using the collected data in order
to identify whether the fit between the manufacturing environment and the planning and control system
is related to the manufacturing performance.
In a theoretical paper, Yamashita et al. (1989) proposed two different approximate solution methods
to minimize product shortages and excesses of inventory for a production line which is operated
according to demand for products. The production line efficiency is measured by the sum of the average
shortage of products and the average inventory of the in-process and finished products in steady state.
The capacities of safety stocks and warehouses are adjusted based on demand. The authors validated
their approaches by comparing their solutions with those obtained by computer simulation.
Por, Stahl and Temesi (1990) developed and presented a DSS for solving production scheduling
problems. The authors characterized the types of production processes that can be handled by the
proposed DSS. The systems' modules, logic and operation are also described in the paper. Their DSS
provides the user with the capability to plan and schedule production in order to satisfy a set of multiple
criteria, including cost, time, or plant utilization.
Verakool and Daily (1988) designed and developed a DSS for production planning and bid prepara-
tion to provide a small job shop manufacturer with an effective day-to-day operations planning System.
Their DSS covered the concerns of a manufacturer to enhance productivity, improve production
scheduling, reduce time for accurate bidding and improve control of material and product inventories.
Duchessi and O'Keefe (1990) discussed the use of a knowledge-based approach to the process of
establishing production rates, work force levels, on-hand inventories and production planning as a viable
alternative to traditional operations research models. The authors describe the production planning
experience in a company using common planning constraints and heuristic procedures, as well as a
prototype of a knowledge-based system for the production planning of one product family.
Wu and Tabucanon (1990) developed a microprocessor based DSS to help medium scale firms make
decisions in a rapidly changing market conditions. Their system consists of three modules: database,
model base, and the interactive dialog, The data base module integrates the advantages of using a
structured data base to computerize manual operations and to provide information. The model base
module is aimed at minimizing inventory cost. A large scale bicriterion integer programming model, for a
real application is solved using a heuristic. The interactive module is designed to allow for easy
user-machine interface. The DSS is based on practical operations of a medium size producer.
In addition to the above generic production planning and decision support systems, several computer-
based systems to plan production of a specific product in a given industry have been reported. Jager et al.
(1989) presented a DSS for planning chemical production of active ingredients in a pharmaceutical
company. Their DSS is based on mixed integer programming (MIP) models to determine the production
quantities of all active ingredients and intermediates for the whole planning horizon. A dialog module
links the MIP models to the production data bases, using a relational type of data base (ADABAS) and a
A.A. Elimam /European Journal of Operational Research 81 (1995) 17-34 19

fourth generation language (NATURAL). The MIP model is optimized using the MPSX 370 Optimizer.
The authors emphasize that the user acceptance of this DSS has been achieved by strict adherence to
two features. First, the system is capable of performing all the planning functions usually available in an
MRP II environment. Second, the DSS applies Operations Research (OR) techniques such as (LP)
without exposing the user to OR jargons or complexities. Fieldhouse and Nuttall (1990) showed that
productivity of paper-mills can be improved by applying computers in production planning. In particular
the authors introduces a Haverly production planning system for the paper industry. In a paper dealing
with production planning in leaf-spring plant, Bhadury and Agarwal (1991) indicated that the nature and
variety of products, the manufacturing process involved, the plant layout, and the existing operating
modes have a considerable impact on the planning system. Moreover they point out to the importance of
designing interactive production planning systems in leaf-spring manufacturing.
In this paper a DSS for planning the production in a US pesticide manufacturing plant is presented.
Even though planning of pesticide production shares many of the salient features of the above
production planning systems, it has many of its own unique characteristics. Since the plant in considera-
tion exports pesticide to several countries, the demand for its products is highly impacted by the
seasonality of the agricultural cycles of those countries. Moreover the plant had to diversify its family of
products and their packaging in order to handle the variety in produce as well as the differences in
agricultural practices in those countries. This would impact the length of production runs, and the
inventory levels. In addition to the seasonality on the demand side, the plant has also to deal with
seasonality on the labor supply and cost side, caused by the annual negotiation with labor union.
Therefore, the production planning problem in the pesticide plant involves several trade-offs covering
the length of the production run, the timing of peak demand in relation to the potential annual increase
in labor cost, the inventory carrying cost versus higher labor cost and the raw material availability versus
cost.
The optimization-based DSS consists of six loosely coupled modules, namely: the Dialog (D), the
Input and Database Update (IDU), the Planning Preprocessor (PP), the Mathematical Programming
Model Generator (MPMG), the Optimization Program (OP) and the Report Writer (RW) modules. Such
DSS has been successfully applied to plan production of pesticide in one of the agricultural products
plants owned by a major US corporation. The paper illustrates such implementation and provides sample
outputs of the production planning DSS.
The paper consists of five major sections. The second section describes the production environment
and presents the problem definition. The production planning system including the mathematical
programming formulation as well as the DSS are given in the third section. System Implementation and
Computational results are discussed in section four. The paper ends with the concluding remarks.

2. Problem description

In this section, the elements of the problem including the products and the production facility as well
as the production planning trade-offs are described. Fig. 1 presents an overview of the agriculture
pesticide production system.

2.1. Product type a n d d e m a n d structure

The plant offers three formulations of pesticide, each can be packaged into several sizes depending on
weight or volume. Demand is specified in terms of formulation and size during each time period.
Because of the product use in protecting agricultural crops in the US and worldwide markets, the
20 A.A. Elimam / European Journal of Operational Research 81 (1995) 17-34

o Availability PRODUCTION FACILITY FORMULA-


- 5 day TION SIZE CUSTOMER I
-7day
4xl 0 o Demand
o Cost _~ 10 G 2xE5
o Seasonal LINE 104 --_on 2x25B o Period ]
Adjustment 223#
==
o Type o 2X25
c 15 G 2X30 INVENTORY"
I 0 Min, q 10KG
o Amounts LINE 165 I o Minimum
I tock
stock
0 Cost 5G 223 O Cost
o Cost I

PRODUCTION PARAMETERS
a Capacity
o Operating mode
o Production rates
o Raw material
consumption rate
o Labor needs

Fig. 1. Pesticide production system.

demand for pesticide is seasonal and fluctuates considerably over the months of the year. In fact the
plant would not be able to satisfy peak demand without resorting to inventory. Because of the
considerable variation in demand, the plant management opted for continuously keeping a safety stock
for each period.

2.2. Production facility

The production facility consists of two production lines, each with its own capacity and production
rate. Each line can be run on a 5 or 7 day operating mode depending on demand, inventory and raw
material availability levels. The production rates will vary on the basis of the operating mode.
Plant Workers are members of a labor union. Their wages vary according to the operating mode of 5
or 7 days per week. Moreover these wages are renegotiated with the Union every year which usually
results in wage increase during the following year. The labor cost follows a piecewise linear function to
reflect economy of scale and shows that the per unit labor cost decreases with the increase in production
quantities (Fig. 2).
The pesticide is made of one primary type of raw material supplied by two major vendors. As it
happens the more expensive vendor can supply larger amounts than the other supplier.

2.3. Planning trade-offs

In producing the agriculture pesticide, the plant faces several decisions in order to weigh the
advantage of the various planning trade-offs. Such decision involves: (1) selection of the optimum
production run length, (2) timing and amount of production in light of demand seasonality, the annual
A.A. Elimam ~European Journal of Operational Research 81 (1995) 17-34 21

bijI = labor costs for pesticide i produced on line j


within grid range I

i .... t ..........

1O0 200 1300 41 O0 10000


Fig. 2. Piecewise linear function for labor cost.

increase in labor cost and the inventory carrying cost and (3) raw material availability versus cost. Fig. 3
illustrates the seasonal variation in demand relative to the expected increases in labor cost.

3. Production planning decision support system

3.1. System design and description

Several steps and guidelines were followed in designing the DSS for planning the production of
pesticide. It all started with the assessment of the Decision Maker information needs. The structured
decision analysis approach was used to identify all the decisions made by the DM. The system was
intended to help the production planner, and in some cases the plant manager, decide on:

• Production and inventory levels of each pesticide formulation and package size required to meet
demands.
• Inventory level of raw material.
• Plant operating mode.
• Annual plant operating Budget.
• Production lines Utilization levels.
The Production planner would like to reach the foregoing decisions while satisfying the plant capacity
and material availability constraints at minimum cost. The requirement to provide a minimum cost
production plan directed our development effort towards an optimization-based DSS.
Moreover, the user needed the capability to quickly perform What-If, Goal-seeking and Sensitivity
analysis studies. The user needed modelling capabilities that can help in making different purpose runs.
22 A.A. Elimam / European Journal of Operational Research 81 (1995) 17-34

720?
640-

560- B i = ith increase in labor cost, %

$
480-

•~
400-

320-
240-
160-

80-
I L 1
(1+Bt) (1+B2) Laborcost
(I+B 1) (1+B2)-

(I+B1)-!
(I+B 1) Labor cost
I
Labor cost

I i I i i i I i i I I I i i I I i I ID
F M A M J J A S O N D J F M A M J J A

Months
Fig. 3. S e a s o n a l d e m a n d a n d a n n u a l a d j u s t m e n t s of l a b o r cost.

For example the DM needed to discourage short production runs. A piecewise linear formulation for the
production cost was used to achieve such need. In the meantime the DM wanted to have the ability to
make approximate and simple runs for quickly exploring various planning and budgeting alternatives.
Therefore a pure linear programming model was also included. The user needed to have access to these
different modeling capabilities without getting exposed to the intricacies of modelling and software
changes. Hence it was decided to develop a computerized model builder that generates the model
components without exposing the user to the intricacies of changing mathematical expressions or
processing data.
The Production planner also needed the ability to manually specify or override part of the production
plan. Accordingly the input mechanism to achieve such capability was introduced into the systems'
design.
The DM spends considerable time collecting and updating information about customer demands,
changes in costs and production lines capacities, among several other data parameters. This pointed to
the need for a separate data base maintenance and update module. It was decided to have a loosely
coupled system modules, which enables the DM to run only the module he needs to use.
The plant uses specific report format for production planning and budgeting. In the interest of saving
re-training time throughout the plant, the DM wanted to have the flexibility to produce similar reports
over different planning horizons and for a specific group of products or facility. A special report writer
module was included to respond to such need.
The plant is part of a major US corporation that invested heavily in establishing an impressive
mainframe computing environment. This, coupled with the fact that the current PC-based computing
technology was not available at the time of developing this application, led to the selection of mainframe
hardware and software as the computing environment for system development.
A.4. Elimam ~European Journal of Operational Research 81 (1995) 17-34 23

I Input& DataBase UpdateModule ] ~ ~ I ReportWriter Module


tlnput parameters I~ ~ [ •I n t e ~ a s e
| | informationwith optimizedplan
i Data updates I~ ,e~ • Producethreetypesof reportsper
• Reporting

Planninq PreprocessorModule / [ MatrixGeneratorModule ] I Optimization


Module
• Determineplanninghorizon& size ~_.~
• I •Gene-~ate ~e~nction ~ . ~
and constraints eOptimiZeplan
modeltoprovideproduction
of model i • Buildtheformulationarraysfor • Storeproductionplaninan arrayforthe
i optimization
module reportwritermodule

Fig. 4. A view of the DSS modules for pesticide production planning.

The DMs' need for flexibility in the input and output modules as well as the ability of these modules
to easily interact with the model builder and the optimization program led the development team to
select the mainframe PLI - based Extended Control Language (ECL) software package. This allowed for
the use of a high level language such as PLI in developing the codes for the input, the data base update,
the model builder and the report writer modules, while being able to easily communicate with the
mathematical programming solution macros available in the IBM MPSX370 software Package. These
computing facilities already existed in the company. The use of existing computing resources helped in:
(1) reducing the system development cost, (2) alleviating the need for extensive user training and (3)
providing regular and reliable system maintenance and user support.
In a nutshell our extensive interviews and interactions with the DM identified the need for an
optimization based production planning model, a separate input and database update program, an easy
to use system dialog and a flexible report writing mechanism. The need for an optimization model led to
the introduction of three separate programs, namely: the planning preprocessor, the model builder and
the optimization modules. Therefore, the optimization-based DSS consists of six modules (Fig. 4),
namely: the Dialog, the Input and Database Update (IDU), the Planning Preprocessor (PP), the
Mathematical Programming Model Generator (MPMG), the Optimization (OP), and the Report Writer
(RW) modules.
Inputs to the DSS include all demand data, capacities and operating costs of production lines, cost
and availability of raw material from various sources, production rates and minimum ending inventory
levels.
The output of the system contains the amount of production from each formulation/size combination,
utilization of facilities and the associated variable costs.
Processing starts by updating the database, as needed. The IDU module can be run as stand alone or
as a starting point for the rest of the system. The second step involves running the PP which retrieves
information from the database and merges it with the input records required to fill the arrays needed for
the MPMG module. These arrays are passed to the MPMG which builds the optimization input array in
24 A.A. Elirnam / European Journal of Operational Research 81 (1995) 17-34

a BCD format, using the Extended Control Language (ECL) for the MPSX370. The OP module, which is
an ECL MPSX program uses this array to optimize the production plan. Finally the RW produces output
reports for up to eighteen month production plans. The description and major functions of each module
is presented next.

3.2. System modules

Dialog (D). This is a menu driven module and it allows the production planner to interact with the
other modules of the DSS. It guides him in making requests from, or giving commands to, the system and
receiving the response. In order to save the production planner the burden of specifying technically
oriented commands, a plain english passive style menu selection was adopted for interacting with the
system.
Input and Database Update (IDU). This program updates the contents of the planning database. It
contains 12 types of records including products code/name, raw material code/name, supplier
code/name, facility code/name, demand data by product code, minimum stock requirement, raw
material consumption, production rates, raw material cost and labor cost. The user has the option of
deleting, adding, replacing or updating a record in the database.
Report Writer (RW). The report writer uses the output of the OP module to generate three types of
reports:

(i) Monthly production plan for each formulation which contains:


Starting inventory.
-

- Production levels on each line in MLBS and in days.


- Total production in MLBS/days.
- Ending Inventory.
(ii) Monthly utilization report for each line which contains:
- Production levels per product/line in MLBS and in days.
- Capacity in number of days.
- Mode of operation.
- Total production on each line in MLBS and in days.
(iii) Monthly cost report which consists of:
- Cost of raw material by supplier, and total.
- Cost of labor by line, and total.

Because of the seasonality of demand and the annual increase in labor cost, the RW is designed to
report an eighteen month production plan, by default. However the user has the option to specify any
length of planning horizon. As part of the production control process, the user can supply the observed
data for any part of the plan that has already passed. This part will be included in the entire 18 month
production plan report.
Planning Preprocessor (PP). The planning preprocessor module uses the database along with other
user specified inputs to prepare the basic data needed for the MPMG module.
The user specified static data include:
-Starting period of the production plan.
- Number of periods in the plan.
- Discounting factor.
-~ Selection of pure LP or Separable model.
- Initial pesticide inventory levels for each formulation/size.
A.A. Elimam~European Journal of OperationalResearch 81 (1995) 17-34 25

- Initial raw material inventory levels.


- Bounds on production line capacities and operating m o d e for all periods.
- Steady state production rates.
- Raw material availability.
Output of the PP module includes several arrays which are supplied to the M P M G in addition to an
echo check summary on input data.
Optimization Program (OP). This module is written in the E C L of MPSX370 to find the least cost
production plan. The control p r o g r a m uses a sequence of M P S X macros to come up with an initial
solution, then improves on that solution until an optimum is reached. The optimized plan is loaded into a
structure which is passed on to the R W module, in the same order the variables and the constraints were
supplied in the M P M G module.
Mathematical Programming Matrix G e n e r a t o r (MPMG). This matrix generator receives all its input
requirements from the PP module in external arrays. T h e n it builds the BCD file i n a free form
acceptable to MPSX370. The inputs to this module include:
- N u m b e r of periods.
- N u m b e r of pesticide formulations.
- Separable or LP model switch.
N u m b e r of production lines.
-

- L a b o r cost within each quantity grid for each p r o d u c t / l i n e .


Material cost by supplier.
-

- Material consumption for each pesticide formulation.


- Production rate for each p r o d u c t / l i n e .
- Labor cost adjustment factor.
Discount Rate.
-

Monthly Sales Demands.


-

Monthly minimum inventory levels for each formulation.


-

- Initial inventory levels for each formulation.


- U p p e r bounds on raw material availability from each supplier.
- U p p e r bounds on the capacity of each line in n u m b e r of days per month.
- If needed, u p p e r bounds on the special variables in a separable programming m o d e l
The above input data are used to build the model in a ' R O W ' section followed by ' C O L U M N ' , ' R H S '
(right hand side) and finally a ' B O U N D ' sections, respectively. These sections are read by the M P S X
optimization control program. The structure of the model generated by the M P M G module is presented
next.

3.3. Mathematical programming model


Notations are defined first followed by statement of the mathematical model.

a) Notations
The mathematical symbols, to be used in the model, are defined under two major categories: decision
variables and input parameters.
Decision variables:
Xij(t) = A m o u n t of pesticide i, produced on line j during time period t, MLBS.
Ii(t) = Inventory level of pesticide i at the end of time period t, MLBS.
Yijt( t ) = Incremental variable, representing the amount of pesticide i, produced within grid l on line
j" during time period t, MLBS.
u/(t) = Surplus in inventory level of pesticide i over the minimum stock required by the end of time
period t, MLBS.
26 AM. Elimam / European Journal o f Operational Research 81 (1995) 17-34

V~(t) --- Shortage in inventory level of pesticide i, below minimum stock level by the end of time
period t, MLBS.
Skp(t) = Amount of raw material k, supplied by vendor p during time period t, MLBS.
W:(t) = Under-utilization of production line j, during time period t, days.
Input parameters:
Di(t) --- Demand for pesticide i during time period t, MLBS.
Rk(t) -- Inventory level of raw material k, by the end of time period t, MLBS.
GUB l = Breaking point for piecewise linear segment l representing pesticide production, MLBS.
£1ki -~- Amount of raw material k used for pesticide i, MLBS/MLBS.
Bi( t ) = Minimum buffer stock of pesticide i by the end of time period t, MLBS.
Cj(t) = Capacity of production line j during time period t, days.
PRij = Production rate of pesticide i on line j, MLBS/day.
bijl(t) = Labor cost for pesticide i produced on line j within grid l during time period t, $/MLBS.
rkp(t) = cost of raw material k from supplier p during time period t, $/MLBS.
CUB:(t) = Upper bound on capacity of production line j during time period t, MLBS.
CLBj(t) = Lower bound on capacity of production line j during time period t, MLBS.
RUPkp(t) = Availability of raw material k from supplier p during time period t, MLBS.
Monthly inventory holding cost of pesticide i, $/MLBS.
PS i = Penalty due to shortage of pesticide i, $/MLBS,
Ek = Monthly inventory holding cost of raw material k, $/MLBS.
PU: = Penalty due to under-utilization of production line j, $/MLBS.
h = Monthly interest rate, %.

b) Model formulation
I. Constraints
(i) Pesticide inventory balance:

Ii(t ) = I i ( t - 1) + ~ Xiy(t ) +Di(t), i=l,...,m, t=l ..... r. (1)


j=1
(ii) Raw material inventory balance:
P m

Rk(t ) = R k ( t - 1 ) + ~] Skp(t ) - ~_~akiXii(t), k = l . . . . . K, t=l,...,r. (2)


p=l j=l i=l

(iii) Pesticide desired minimum stock:


li(t)-Bi(t)=Ui(t)-Vi(t), i = 1 . . . . ,m, t = 1 . . . . ,T. (3)
(iv) Production lines capacity constraints:
Actual Production/< Capacityt , (4)

(5)
j~l

However one might introduce a goal constraint to reflect the under-utilization of capacity.
Explicitly

Q(t) - Xi:(t) = ~.(t), j= 1,...,n, t= 1,...,r, (6)


j=l

where g~:(t) is the slack in the capacity of production line j during time period t.
A.A. Elimam ~European Journal of Operational Research 81 (1995) 17-34 27

Moreover we can allow the capacity, Cj(t), to vary within certain range. In this case, Cj(t)
becomes a decision variable instead of an input parameter.
Therefore
CLBj.(t) _< Cj(t) < CUBs(t), j = 1 . . . . . n, t = I,...,T. (7)
(v) Piecewise linear representation of pesticide production levels:
L
Xiy(t)-EY~jt(t)=O, i = 1 . . . . . m, j = l .... ,n, t=l,...,T, (8)
t=l

and
Y,-jt(t) -<GUBt, i = 1. . . . , m , j= l,...,n, I= I , . . . , L , t= l , . . . , T . (9)
(vi) Limit on raw material availability;
S~p(t)<RUBkp(t), k = l . . . . . K, p = l . . . . . P, t = l . . . . ,T. (10)
(vii) Non-negativity constraints on all the decision variables defined above.
II. Objective function
The optimization criterion includes two major elements: actual costs and Penalties. The first deals
with the actual cost of production and it consists of the labor, raw material, and inventory holding costs.
The labor cost decreases, in a piecewise linear fashion as the amount of production increases. The raw
material cost depends on the source of material. Inventory holding costs are computed for both raw
material and pesticide. The labor, raw material, and inventory holding costs are discounted at a rate h.
Therefore the objective function is given by the following expression:
r 1
(1 + h) t-x [(Labor Cost)t + (Pesticide Inventory Holding Cost)t
t=l

+ (Raw Material Inventory Holding Cost)t + (Raw Material Cost)t]


T
+ ~ [(Minimum Stock Violation Penalty)t + (Capacity Under-Utilization Penalty)t]. (11)
t=l

This can be stated by the following mathematical expression:

E ,l+;_;t-, bii,(t)Yij,(t) + E E rkp(t)Skp(t) + E H i l i ( t ) + E EkRk(t)


t=l /=1 i=1]=1 k=l p=l i=1 k=l

+ E esi(t) (t) + E euj (t . (12)


t=l j=l ]

The last element, contained in the square brackets, of the objective function represents the penalties
accrued due to: (i) violating minimum stock levels of pesticide i, and (ii) under-utilizing the production
capacity of line j.
The preceding model consists of T[3m + n + (1 + L)nm + KP] variables, T[2m + 2n + mnL + K]
constraints and T[2n + mnL + KP] bounds. The formulation is mostly linear, with the exception of the
piecewise linear representation of pesticide production levels provided by constraints (8) and (9) as well
as the first term in expression (11). This model has been incorporated into the above DSS modules. In
particular the PP, MPMG, OP, and the RW modules have been implemented to carry out aU the
processes required to develop, optimize and report the resulting output of the above model.
28 AM. Elimam / European Journal of Operational Research 81 (1995) 17-34

4. Production planning DSS implementation

4.1. System development

The system has been developed and implemented on an IBM mainframe to determine the least cost
pesticide production plan for a major US petrochemical corporation. The OP module made use of the
existing MPSX370 software package for optimizing the mathematical programming model. All other
modules have been developed specifically for this application. The system has been built with interface
files among modules. Such loose coupling allowed the user to run each module independently or in
concert. Clearly a module cannot be run unless its input requirements were available.
The development team relied heavily on extensive interviews and interactions with the production
planner and the plant manager. Menus, input screens and output reports were designed and accepted by
the user before proceeding with implementation. System input screens are illustrated in Figs. 5 and 6.
The system module selection of the Dialog program is given in Fig. 5. Product demand updating sample
is presented in Fig. 6. The development of such system required resources from different departments in
the company. Management support was essential in receiving the cooperation of the departments in
control of the needed resources. In fact the plant management and user support and cooperation played
a major role in the successful development and implementation of this DSS.
Each completed module was subjected to independent testing before it was integrated into the DSS.
The user supplied benchmark test files that were used extensively in validating results. The completed
system was tested for more than three months before it was adopted as the production planning and
operational budgeting tool of the plant. During this testing period, the systems' performance was
monitored and compared with the conventional production planning approach. Such testing period
provided the production planner with the opportunity to build up confidence in the system and its
capabilities. One of the problems that was encountered during testing is the technical nature of error
messages printed by the LP optimization program. For example, the user would receive an 'infeasible'
error message that might be hard to diagnose. Therefore a special effort was made to classify all
potential causes of infeasibility that could surface in the model. The user would only see pragmatic
interpretations of these causes along with suggestions to remove them.

RECORD TYPE (select 1-12 and hit enter):

PRODUCT CODE (P1._P12) [ ~

DEMAND DURING Year ~ Month


C-r-q
AMOUNT:
1 INPUT NEW DATA OR UPDATE EXISTING DATA
2 GENERATE REPORTS FOR EXISTING PRODUCTION PLAN MORE DEMAND UPDATE (Y/N)?
(continue if 'Y', otherwise terminate)
3 GENERATE A NEW PRODUCTION PLAN FOR EXISTING DATABASE
4 GENERATE A NEW PRODUCTION PLAN FOR A NEW DATABASE
5 EXPLORE IMPACT OF INPUT CHANGES ON EXISTING PLAN
SAME PRODUCT-DIFFERENT PERIOD (Y/N)?
(If 'Y' provide same screen - blank Period & Amounts)
Vrq
USER SELECTION I I SAME PERIOD - DIFFERENT PRODUCT (Y/N)?
(select number or highlights then hit "ENTER" to run option) (If 'Y' provide same screen - blank Product code & Amount)

Fig. 5. Agricultural Pesticide DSS Module Selection Input Screen.

Fig. 6. Sample Input Screen for Product Demand Update.


A,4. Elimam~EuropeanJournalof OperationalResearch81 (1995)17-34 29

4.2. Computational results

In this section the results of the pesticide production planning DSS is illustrated by developing a
production plan to satisfy demand for eight formulation/size combinations over 18 months. Because of
past commitments, the production planner decided to specify production levels for the first four months.
The DSS provides three types of reports, namely:
(i) Monthly production Report for each product.
(ii) Monthly Utilization plan of each production line.
(iii) Monthly raw material and labor cost.
These reports are illustrated in Tables 1, 2, and 3. The results in Table 1 show an 18 month
production plan for each pesticide formulation. This plan consists of a manually predetermined
production plan for the first four months and the optimized plan for the last 14 months. The utilization
plan of each production line is also illustrated in Table 2. The variable operating costs of the plant are
shown in Table 3. Careful scrutiny of the production plan presented in the three tables points to the
following observations
• The system tends to run large batches of production. For some types of pesticide formulations, full
month production runs were scheduled. This is attributed to the use of piecewise decreasing labor
cost.
• Because of the inventory holding cost, the annual increase in labor cost seems to have limited
impact on the production schedule.
• Production line 104 is fully utilized during the period of June through December. On the other
hand, production line 165 had to be run overtime during the months of February, March, and April.
For the remaining months, it was run for a full one-shift capacity. Line 165 has higher utilization
rates because of its improved and modem technology as well as its higher efficiency.
• The variable cost table indicates that the raw material cost dominated the labor cost throughout the
18 month planning horizon.

4.3. Benefits of pesticide production planning DSS

The system has been in use for more than five years in one of the agriculture products plants of a
major US petrochemical Corporation. In order to assess its impact on the plant operation, several
measures of performance were monitored, before and after adopting this DSS. These measures include
utilization of production lines, total unit cost in S/unit, labor cost, average inventory level, and the
average production run size. These measures of performance were obse~wed for a period of one year
after applying the system. Historical values for these measures were used to compute annual values of
the plant's performance.
Table 4 summarizes the percent improvement in these measures. The table shows that the system
helped in realizing higher utilization of production lines, lower labor and production cost per unit,
reduced inventory levels and a significant increase in production run size. The latter has been achieved
due to the fact that the production cost is represented by a piecewise linear function to reflect economy
of scale. The overall reduction in cost is attributed to the improved synchronization between seasonality
of demand, labor cost increases, and material availability.

4.4. Other functions of the DSS

In addition to realizing major improvements in the performance measures, presented earlier, the
system has proved to be beneficial in several other uses, namely:
30 A . 4 . E l i m a m / European Journal o f Operational Research 81 (1995) 1 7 - 3 4

o. o. c~ o.
0
¢.q

o. o o. o
o 0 0 0 ~"~o o,~ OC~
o

0 0
O 0 0

0 ~ 0
0 ~ ~ 0

0
0 ~ ~ 0 0 ~ r¢3
~ t~
t.-i
0
0 0 o eel

C~
0 0 o

0
0 o (.,1 t,¢3

0 ~ o 0
0
e¢l ~ r'..

0
0 ~ ~ 0 0

t.-,i
C~a 0 t~ o
0 0

t"-I
C~
0 o ~2~ ~ t-.,i
t".l
p.. c.,1
t~
c~ o. o. q
~ 0

c~ o. o.
0

c~ o. o
c~c5 ~c~

o 0. o.
0 t-¢3 o 0 oo ~ ~ o

o o. 0. ~ o
t".l ~

e~ t¢3

v~

Z Z
Z Z
Z ~ 0 0 © Z Z
0 o o o o ~ ~ o
Z
r~

Z
m ~ r~ N
A.A. Elimam ~European Journal of Operational Research 81 (1995) 17-34 31

o. ~. o. o. ~. o. o.
~'~ O 0 ~ 0 O 0 O 0 O 0 O 0 O 0 0 0

o. o o o o. o. o 0
0 0

o. o o o. o o. o. 0
0 0

~Z
o o o. o. o. o. o. 0

o o o. o o o.
o~5 oo oo oo o~ oo

0 0 o 0 o

0 o o o
o o o o

0 0 q o o
o 0 o o
0 ~D e'q

kO 0 0
0 o 0 0

0 0 o. o o 0
m~ o 0 o 0 0 o

2
o. o o o
~D O 0 O 0 0 0
1¢3 cN
< t'xl

0 t~ 0 0 o
o 0 0 o
t~

0 o cq ~D o t".l
Z 0 o
0 t'xl

cq 0 0
t~ o ~ O o
0 t"l

~Z
o o. o. o o. o o. o. o.
< t~

o. ~. o. o. o. o. ~. o. o.
O 0 O 0 O 0 O 0 O 0 O 0 O 0 O 0

0
..~

o o. ~ o. o o. ~. o.

N~ N
t¢3

0
32 A.A. Elimam / European Journal of Operational Research 81 (1995) 17-34

c~

¢-

>

b-
(Z
v-~ t'q

o ~-t".lxg)
¢Z,

7~

<

Z
o
e~

o
AM. Elimam / European Journal of Operational Research 81 (1995) 17-34 33

Table 4
Improvements in the annual measures of performance for the pesticide production planning DSS
Annual measure of performance Percent improvement after system use
• Production line utilization
- Line 104 4
- Line 165 3.2
• Reduction in cost
- Total 5
- Labor cost 2
• Reduction in average inventory level
- Raw material 1
-. Pesticide 1.7
• Average increase in production run size " 11.4

• Provide guidelines for labor cost increases during annual negotiations with the labor union. By
conducting sensitivity analysis on impact of labor cost, bijt(t), on the resulting production plan and
the associated measures of performance, the company was able to develop a range for the allowable
increase in labor cost without jeopardizing the company's financial viability.
• Develop purchasing policies to ensure raw material availability at minimum cost. The purchasing
d e p a r t m e n t used the system's output to synchronize its purchase orders with the actual plant needs
without needlessly raising inventory levels.
• Assist in devising the plant's annual budget on the basis of the least cost production plan. The
report writer module included a costing program for calculating the plants' budget, based on the
developed pesticide production plan. In fact this feature has proved to be very useful to the
budgeting and cost control department. It led to pursuing further system development to allow the
user to update and compare actual versus planned for each cost item as well as the total.
• Provide the mechanism to achieve the production control function. The DSS included the capability
to receive feedback on actual production levels and compare these levels with the planned ones.
Major deviations were diagnosed using the system's outputs. The impact of these deviations is
reflected on future planning periods by varying input p a r a m e t e r s to the DSS for these periods.
• H e l p structure the data base for the plant including consistent recording of demand for pesticide,
production line capacities and utilizations, material purchases by supplier, material consumption
factors, labor costs and their annual adjustments. The model requirements and outputs provided the
structure required for designing, establishing, and maintaining a consistent and reliable data base
for various p a r a m e t e r s needed by the pesticide production facility.

5. Concluding remarks

This p a p e r dealt with the developm~ent of a mathematical programming-based DSS for the production
planning of agriculture pesticide. T h e model includes a piecewise-linear labor cost function. Moreover, it
allows for the annual adjustment of labor costs. T h e objective function takes into account the inventory
holding cost. The constraints allow for variable plant capacity based on the selected operating m o d e as
well as the production rates for each pesticide formulation. Inventory balance constraints, and minimum
stock levels for pesticides and raw material are also included into the model.
The mathematical progralmning model has been incorporated into a DSS which consists of six
modules, namely: dialog, input and database update, planning preprocessor, mathematical p r o g r a m m i n g
matrix generator, optimization program, and the report writer. The DSS was developed using PLI and
the extended control language of I B M ' s MPSX370. Tile system has b e e n successfully implemented in a
southern US pesticide manufacturing plant.
34 A.A. Elimam / European Journal of Operational Research 81 (1995) 17-34

The development and application of the DSS for pesticide production planning led to the following
conclusions:
(i) The system helped plant management in providing a more positive and reliable response to the
fluctuating demand for pesticides, particularly from developing countries, at minimum cost.
(ii) The model has also eliminated short production runs of various pesticide formulations.
(iii) The plant was able to achieve the least cost balance among seasonal demand, the increase in labor
cost, and availability of raw material from lower cost suppliers.
(iv) The system provided management with an accurate tool for cost accounting and control, based on
planned and actual plant operations.
In a nutshell, the Pesticide production planning DSS has proved to be of merit and provided a reliable
and effective tool in planning and controlling plant operations. The implementation of this system has
enabled the plant to meet fluctuating seasonal demand for its products while minimizing production and
inventory costs as well as product changeovers.

References

Bahdury, B., and Agarwal, S,K. (1991), "Development of a production planning and control System for a leaf-spring Manufacturing
Facility", Production and Inventory Management Journal 32/2, 1-6.
Duchessi, P., and O'Keefe, R.M. (1990), "Knowledge-based approach to production planning", Journal of the OperationalResearch
Society 41/5, 377-390.
Fieldhouse, M., and Nuttall, G. (1990), "Production planning with computer assistance", Paper Technology and Industry 31/10,
22-24.
Higle, J.L. (1991), "Production planning with discounting and stochastic demands", European Journal of Operational Research
50/3, 257-265.
International Business Machines (IBM), "Mathematical Programming System Extended/370 (MPSX370. 1979 Program Reference
Manual", SH19-1091, IBM Corporation, White Plains, NY, USA.
Jager, K., Peernoller, W., and Rohocle, M. (1989), "A Decision Support System for planning chemical production of actice
ingredients in a pharmaceutical company", Engineering Costs and Production Economics 17, 377-387.
Mattsson, L., Olhager, J., Ovrin, P., and Rapp, B. (1989), "Computerization of manufacturing planning and control systems",
Engineering Costs and Production Economics 17/1-4, 71-88.
Ovrin, P., Olhager, J., and Rapp, B. (1990), "Computer-based manufacturing planning and control systems in small companies",
European Journal of Operational Research 45/1, 106-112.
Por, A., Stahl, J., and Ternesi, J. (1990), "Decision Support System for production control, multiple criteria decision making in
practice", Engineering Costs and Production Economics 20/2, 213-218.
Sirdharan, V., Newman, W.R., Chapman, S.M., and Jacobs, R.A. (1989), "Evaluating the interaction between the manufacturing
environment and planning and control systems", Journal of Manufacturing Systems 8/4, 329-332.
Verakool, V., and Daily, M. (1988), "Decision Support System for production planning and bid preparation for a small job shop
manufacturer", Computers and Industrial Engineering 15, 162-165.
Wu, D.S., and Tabucanon, M.T. (1990), "Multiple criteria Decision Support System for production management", Engineering
Costs and Production Economics 20/2, 203-212.
Yamshita, H., Nakazawa, K., Nagasuka, H., and Suzuki, S. (1989), "Modelling and analysis of a production line operating according
to demand", JSME International Journal, Series 3: Vibration, Control Engineering, Engineeringfor Industry 32/2, 491-497.

You might also like