Sie sind auf Seite 1von 15

Available online at www.sciencedirect.

com

Applied Mathematical Modelling 33 (2009) 12131227


www.elsevier.com/locate/apm

Integration of generative and evaluative models for production


scheduling of lube oil plants in a petroleum renery
John Nelson Abraham *, Kavuri Suryaprakasa Rao
Department of Industrial Engineering, College of Engineering, Guindy, Anna University, Chennai 600 025, India
Received 1 September 2005; received in revised form 1 January 2008; accepted 10 January 2008
Available online 1 February 2008

Abstract
A study was made on the existing practices of production planning, scheduling and prevailing constraints in the six
plants of a lube oil section in a petroleum renery. Based on the data collected from these plants, some generative and
evaluative models were developed. The generative models developed were ow network optimisation (FNO) model and
binary integer linear programming (BILP) model. The evaluative model developed was simulation. The optimal results
obtained from the generative model were fed to the evaluative model to derive the measure of performance. This integration of generative and evaluative models oers an opportunity for better understanding of the subsystem and appropriate
decision making.
2008 Elsevier Inc. All rights reserved.
Keywords: Production scheduling; Generative models; Evaluative models; Petroleum renery; Lube oil products

1. Introduction
That the oil rening industry is facing a condition characterized by decreasing prot margins due to surplus
rening capacity and increasing oil prices is a well-known fact. Simultaneously market competition and rigorous environmental regulations are compelling the industry to make wide ranging modications in its operations. As a result, there is no renery nowadays that does not use process optimization tools to improve
business results. Such tools are applied widely in production planning and scheduling. Moro [1] stated that
many benets have already been reaped due to the use of these tools, but there is a lot more to be done to
achieve their full potential, because they still exhibit many weaknesses. Pinto et al. [2] studied planning and
scheduling models for renery operations, which include non-linear planning models for renery production
and mixed integer optimization models for scheduling. They also analyzed and observed that optimization of
the production units did not achieve the global economic optimization of the plant. Usually the objectives of
the individual units were conicting and thus contributed to a sub-optimal, and many times infeasible overall

Corresponding author.
E-mail address: johnan@redimail.com (J.N. Abraham).

0307-904X/$ - see front matter 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2008.01.012

1214

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

operations. Katayama [3] proposed a production planning procedure for multi-item continuous production
and an optimal product mix using integer linear programming in a petrochemical plant. For a production
planning and scheduling problem in an oil renery, to satisfy the demand while minimizing production cost
and considering storage capacity constraint, Maud et al. [4] proposed a mixed integer linear programming
(MILP) model. Artiba and Riane [5] proposed a multi model system integrated with optimization algorithms,
discrete event simulation and expert system techniques for production planning and scheduling of process
industries. Liza and Pinto [6] proposed a MILP model for optimal scheduling of a lube oil and paran production plant. In this work, the authors compared two mathematical formulations for discrete and continuous
time representations. Jia and Ierapetritou [7] developed a comprehensive mathematical programming model
for scheduling of oil renery operations by decomposing the overall problem spatially into three domains.
Xiong and Nyberg [8] developed a computer integrated manufacturing systems model for a renery system,
and discussed the issues related to production planning and scheduling of the renery. Alle and Pinto [9]
addressed the problem of simultaneous scheduling and optimization of continuous multistage multiproduct
plants with intermediate storage. Potter et al. [10] developed a discrete event simulation model to study the
dispatch performance in a steel processing industry. They built the simulation model using ARENA simulation software package. Chryssolouris et al. [11] presented a simulation based approach for scheduling crude
oils from crude storage tanks to the crude distillation units. Moro and Pinto [12] addressed the problem of
crude oil inventory management of a renery that received several types of oil through a pipeline. Neumann
et al. [13] studied scheduling of continuous, semi-continuous and discontinuous production in process industries where storage facilities and resources like processing units and manpower are limited in capacity. They
applied Branch and Bound method for solving the problem. De Matta and Miller [14] coordinated the shortterm production and inter facility transportation scheduling decisions between a plant that produced intermediate products and a nishing plant that processed the intermediate products into nished goods. They developed an MILP model through which they investigated the variability in plant capacity and costs aecting the
coordination of scheduling decisions as well as the choice of transportation modes and carriers. Floudas and
Lin [15] presented a review of developments in the scheduling of multiproduct/multipurpose batch and continuous processes.
In this paper we are presenting the development of an optimization model for production scheduling in an
oil renery. Lube oil products are produced in the renery by distillation of raw material crude. For production scheduling operations of lube oil section in the renery, a suitable mathematical programming model has
been developed by integrating the operations of the plants. An appropriate simulation model has been proposed to take care of the random events in the lube plants. By integrating optimization and simulation models,
the generated production schedule has been evaluated for optimal performance. This paper is organized as
follows: The process details of the case study organization are presented in Section 2. The framework of
the proposed production scheduling model is presented in Section 3. The mathematical model of the ow network optimization (FNO) model and binary integer linear programming (BILP) model for production scheduling are presented in Section 4. The derivation of production schedule from the results of the BILP model is
presented in Section 5. Integration of generative and evaluative models is presented in Section 6. Validation of
the simulation model is presented in Section 7. Discussion and conclusions are presented in Section 8.

2. Case study
This study was conducted in a public owned petroleum renery in India, which has a capacity of processing
9.5 million metric tonnes per annum of crude oil and producing a variety of petroleum products. When the
raw crude oil is processed in the renery, it produces 15 dierent products, namely, sulphur, hexane, naptha,
liqueed petroleum gas, petrol, aviation turbine fuel, kerosene, linear alkaline benzene feed stock, diesel, lube
base stocks, industrial fuel, asphalt, feed stocks for mex-poly butenes, propylene and paran wax. Of these 15
products produced by the renery, lube based stocks have a higher market value and contribute a major share
of revenue for the renery. The raw crude oil is processed through four stages, namely, distillation, ranation,
dewaxing and hydronishing to produce lube based stocks. The lube based stocks consist of six products,
namely, spindle oil (SP), light neutral (LN), inter neutral (IN), high neutral (HN), 500 neutral (500 N) and

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

1215

bright neutral (BN). These six grades are categorized on the basis of their viscosity levels. Some of the uses of
lube oil are for corrosion protection, heat removal and lubrication.
2.1. Process ow description
The process ow diagram of the lube oil production is shown in Fig. 1. The crude oil is stored in tanks in
the renery premises. Approximately 40% of this crude ow of 400 m3/h is used for processing lube oil products while the rest is used for processing other products like kerosene, petrol, diesel, etc.
2.1.1. Distillation process separation of lube based oils
The rst stage is the distillation process. In this stage, 40% of the total inow of crude oil enters the crude
distillation unit and undergoes a distillate separation process. The lube based oils such as spindle oil (SP), light
neutral (LN), inter neutral (IN), high neutral (HN), 500 neutral (500 N) and bright neutral (BN) are separated
and stored in dierent distillation tanks for further processing. The vacuum residue, which forms about 39% of
total inow to the lube oil products, is stored in a separate tank and is processed in the propane deasphalting
unit (PDU). The output of PDU is bright neutral (BN) which is stored in a separate distillation tank.
2.1.2. Ranation processes removal of aromatics to increase viscosity index
The second stage is the ranation process. In this stage, the separated lube oils in the distillate tanks are
further processed in solvent extraction units. The aromatic components are removed from the lube oil with the
help of solvents. The aromatics are mainly removed to increase the viscosity index of the lube oil. The spindle
oil (SP) and light neutral (LN) distillates are processed in furfural extraction units (FEU), where furfural is
used as solvent. The inter neutral (IN), high neutral (HN), 500 neutral (500 N) and bright neutral (BN) distillates are processed in N-methyl-2 pyrrolidone extraction units (NMP), where N-methyl-2 pyrrolidone is used
as a solvent. N-Methyl-2 pyrrolidone is a better solvent when compared to furfural. The outputs of FEU and
NMP are called ranates and are stored in ranate tanks of dierent grades.
2.1.3. Dewaxing process removal of wax to increase the pour point
The third stage is the dewaxing process. In this stage, the ranate oils from the ranate tanks are further
processed in methyl ethyl ketone dewaxing plants (MEK) to remove the wax components that are present in

Fig. 1. Process ow diagram of the lube oil section.

1216

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

the ranate oils. The wax components are removed with the help of methyl ethyl ketone, which is used as a
solvent in the plant. The wax components are mainly removed to increase the pour point of the ranate oils of
dierent grades. Pour point is the minimum temperature at which these lube oils can remain without getting
congealed. The pour points of these lube oils are around 615 C. The outputs of MEK are called dewaxed oils
and are stored in dewaxed tanks of dierent grades.
2.1.4. Hydronishing process improvement of colour
The fourth stage is the hydronishing process. In this stage, the dewaxed oils from the dewaxed tanks are
further processed in hydro nishing units (HFU) to improve the colour of the dewaxed oils. This colour
improvement is done by a catalytic treatment using hydrogen. The outputs from the HFU are called lube
oil based stocks and are stored in nished lube oil tanks for further dispatch to consumption points.
In this paper, for easy identication, the names of the six plants and the six products are designated as
shown in Tables 1 and 2.
2.2. Process constraints
Some of the process constraints in the production of lube oil products are ow and yield rates of the plants,
limits on the capacities of the storage tanks, change over time required to shift from one product to another,
and slop occurrence. These are presented in this section.
2.2.1. Flow and yield rates
All the six plants are operated in the production of lube oil products. Each plant is connected to upstream
and downstream tanks to handle the semi-nished/nished lube oil products. All the lube plants have acceptable limits of ow rate ranges for their operation. Due to the dynamic complexity involved, the production
system behaves dierently for dierent sets of ow rates in each of the plants at any given instance. Hence,
the process engineer in the shop oor should decide on the ow rate parameter to be set for each of the plants
based on the current conditions in the plant. All the six plants in the lube oil section also have yield rates, and
these yield rate values vary with the type of the product processed in each of the plants. The yield rate denotes
the desired quantity of output that comes out of the plant. The remaining quantity is sent out of the plant as
byproduct.

Table 1
Plants of lube oil section
S. no.

Name of the plant

Designated in the paper as

1
2
3
4
5
6

Distillation plant unit


Propane deasphalting unit, PDU
Furfural extraction unit, FEU
N-Methyl-2 pyrrolidone extraction unit, NMP
Methyl ethyl ketone dewaxing plant, MEK
Hydro nishing units, HFU

PL
PL
PL
PL
PL
PL

1
2
3
4
5
6

Table 2
Products of lube oil section
S. no.

Name of the product

Designated in the paper as

1
2
3
4
5
6

Spindle oil (SP)


Light neutral (LN)
Inter neutral (IN)
High neutral (HN)
500 Neutral (500 N)
Bright neutral (BN)

P1
P2
P3
P4
P5
P6

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

1217

2.2.2. Storage tank capacity


A number of up stream and down stream tanks are connected to either side of the six plants as shown
in Fig. 1. Individual tanks are assigned for each one of the six lube oil products both up stream and
down stream. Some of the products are provided with more than one tank to store the material. All
the tanks in the renery are identied by their respective tank numbers. The tanks are calibrated at dierent levels as empty, low, safe, high, full, etc. to identify the current status of the material inside the tank. A
plant can process a particular product only when sucient quantity of that product is available in the up
stream tank and sucient space is available in the down stream tank to accommodate the drained material
from the plant.
2.2.3. Change over time
There are six plants involved in the lube oil processing, out of which one is a distillation plant while the
remaining are conversion plants. Of the available ve conversion plants, PL 2 handles only one product,
namely, P6. PL 3 handles two products, namely, P1 and P2, one at a time. PL 4 handles four products, namely,
P3P6, one at a time. Plants PL 5 and PL 6 can handles all the six products, one at a time. Whenever there is a
product change in any one of these plants, there is a time lag involved in collecting the next product.
2.2.4. Slop occurrence
The operating procedure for plant PL 6 is considered as most complex because plant PL 6 has changeover
guidelines. Based on the chemical properties of the products, a pattern is formulated which serves as a guideline in deciding the next product for processing in plant PL 6. Whenever these changeover guidelines are violated, an interface product called slop is produced.
3. Frame work of the proposed production scheduling model
The process ow diagram of lube oil section given in Fig. 1 is taken as the basis for developing a maximal
ow network problem. Fig. 2 represents the ow network model of the lube oil section.
The squares and rectangles shown in the network represent the six lube plants as well as the source and sink
nodes. Each plant is connected to upstream and downstream tanks. The cylinders shown in the network represent the lube tanks of dierent capacities. The levels in various tanks-both upstream and down stream-vary
with the time of operation of the plants. The tanks and plants are the nodes of the network, and these nodes
are connected by arcs. The ow across each arc is represented by upper and lower limits. The triangles shown
in the network represent the byproducts produced by the lube oil plants.
The proposed production scheduling model consists of ow network optimisation (FNO) model and binary
integer linear programming (BILP) model.
3.1. Assumptions in the operation of the FNO and BILP models
The scheduling system proposed in this study is operated under the following assumptions:
(i)
(ii)
(iii)
(iv)
(v)

Product to be produced is selected by the user.


At any given point of time same product is processed in all the stages of the network.
All tanks are assumed to have some quantity at the start of each run.
The operating time of all plants is dierent for each run.
Change over from one product to another is possible only at the end of each run.

3.2. Objectives of FNO and BILP models


(i) The objective of FNO model is to nd the optimum ow rate across each arc in the ow network and to
nd the maximal ow permissible in the network.
(ii) The objective of BILP model is to maximize the throughput while optimizing the time of operation of
each plant.

1218

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

Fig. 2. Flow network representation of the lube oil section.

4. Mathematical model of the ow network


The nomenclature used in the proposed mathematical model is as follows:
Indices
i
j
o
k
n
r

plant number
tank number
source
byproduct
sink
run (iteration) number

Variables
f
ow/rundown rate
t
duration of ow
ow from source o to plant i
fo,i
fi,,j
ow from plant i to tank j
ow from tank j to plant i
fj,i
ow from plant i to byproduct k
fi,k
ow from tank j to sink n
fj,n
ow from sink n to source o
fn,o
time of ow from source o to plant i
to,i
time of ow from plant i to tank j
ti,j
time of ow from tank j to plant i
tj,i

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

1219

ti,k
time of operation of plant i for by product k
time of ow from tank j to sink n
tj,n
bl, blm, blmn binary variables used for selection of same product in both upstream and down stream tanks j
tank level at the start of the run
TPJ
tank level at the end of the run
TCJ
4.1. Flow network optimization model
Objective function:
Maximising the ow through the system:
Max

f n;o

Subject to
Flow balance equation for a node represented by each plant:
Plant-1 (PL 1) handles all six products (P1P6) simultaneously
fo;i1  to;i1

6
X

fi1;j  ti1;j fi1;k1  ti1;k1 :

j1

Plant-2 (PL 2) handles one product, P6


fj6;i2  tj6;i2 fi2;j7  ti2;j7 fi2;k2  ti2;k2 :

Plant-3 (PL 3) handles two products, P1 and P2.


Flow balance for each path in PL 3 is represented by
b1 fj1;i3  tj1;i3 b11 fi3;j8  ti3;j8 fi3;k3  ti3;k3 :

4:1

Tanks j1 and j8 contain semi-nished product P1


b2 fj2;i3  tj2;i3 b21 fi3;j9  ti3;j9 fi3;k3  ti3;k3 :

4:2

Tanks j2 and j9 contain semi-nished product P2


b1 b11 ;

4:3

b2 b21 ;

4:4

b1 b2 1;

4:5

b11 b21 1;

4:6

b1; b11; b2; b21; 2 f0; 1g:


Similarly, ow balance equations for the plants PL 4PL 6 are derived.
Source and sink nodes are connected
fn;o  fo;i1 0:

Limits of arcs in the network for the source node


380 6 fo;i1 6 400:

6:1

In plant-1 (PL 1)
15 6 fi1;j1 6 19;

6:2

10 6 fi1;j2 6 17;

6:3

1220

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

12 6 fi1;j3 6 24;
05 6 fi1;j4 6 11;

6:4
6:5

16 6 fi1;j5 6 22;
77 6 fi1;j6 6 83;

6:6
6:7

Limits on byproduct of PL 1
6:8

fi1;k1 P 234:
In plant-2 (PL 2)

6:9

35 6 fj6;i2 6 60;
Quantity produced based on yield rate
fi2;j7 0:2fj 6;i2

6:10

Limits on byproduct of PL 2
6:11

28 6 fi2;k2 6 48:
Similarly the limits for the arcs associated with plants PL 3PL 6 are derived.
Sink node
20 6 fj20;n 6 200;

7:1

20 6 fj21;n 6 200;

7:2

20 6 fj22;n 6 200;

7:3

20 6 fj23;n 6 200;

7:4

20 6 fj24;n 6 200;

7:5

20 6 fj25;n 6 200:

7:6

Arcs connected to sink node


fn;o

25
X

fj;n  tj;n :

j20

4.1.1. Results from FNO model


The FNO model described above integrates all the subsystems (plants) through the ow network. This
FNO model is solved by using LINDO software package. The optimal ow rates across various arcs of the
network and the maximal ow possible through the network are obtained from the solution. The optimal ow
rates obtained from the FNO model are presented in Table 3. The FNO model is extended by incorporating
the time variable representing time of operation for each plant to satisfy the demand for the products and is
referred as BILP model. The optimal ow rates obtained from the FNO model are fed into the BILP model.
The details of the BILP model are presented in Section 4.2.
Table 3
Optimal ow rates obtained from FNO model solutions in m3/h

P1
P2
P3
P4
P5
P6

PL 1

PL 2

PL 3

PL 4

PL 5

PL 6

19
17
24
11
22
83

60

30
30

60
60
60
60

80
80
80
80
80
80

60
60
60
60
60
60

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

1221

4.2. Binary integer linear programming (BILP) model


Objective function
Maximizing the throughput of the system:
Maximize

25
X

fj;n  tj;n

j20

Subject to
Eqs. (2)(5) and ow balancing equations for the plants PL 4PL 6 of FNO model are derived.
Resetting tank levels at the beginning of each run
TP j;r TC j;r1

for j 1 to 25;

10

where r: run number; TPj: tank level of j at the start of the run; TCj: tank level of j at the end of the run.
Limits on tank levels
1108 6 TC 1 6 7380:

11

The above equation represents the upper and lower limits of the (T1 + T11) tanks. Similarly the upper and
lower limits for the remaining 24 tanks in the lube oil section are derived.
Updating tank levels (current level = previous level + in ow  out ow)
TC 1 TP 1 fi1;j1  ti1;j1  fj1;i3  tj1;i3 :

12

The above equation represents the current level of the (T1 + T11) tanks. It is derived by incrementing the inow and deducting the outow from previous level of the tank. Similarly, the updating equations of the
remaining 24 tanks in the lube oil section are derived.
Check the time of operation for each product type at PL 3 using binary variables
t112  b1  t1121  b11 0;
t212  b2  t2121  b21 0;

13:1
13:2

b1; b11; b2; b21; 2 f0; 1g:


Similar balancing equations are derived for the plants PL 4PL 6 which handle only one product at a time.
Check the time of operation for each plant:
Plant-1 (PL 1)
t1  t101 0;

14:1

t1  t201 0;
t1  t301 0;

14:2
14:3

t1  t401 0;
t1  t501 0;

14:4
14:5

t1  t601 0:

14:6

Similar balancing equations are derived for the remaining ve plants PL 2PL 6.
The above BILP model was run using LINDO software, keeping a nite time period as the planning horizon.
The optimal time of operation of each plant obtained from the solution of the BILP model is presented in Table 4.
4.2.1. Results from BILP model
The BILP model is run by selecting any one product out of the six products. In the second stage of the ow
network, there are two plants, PL 3 and PL 4, of which PL 3 is capable of handling two products, and PL 4 is
capable of handling four products. Based on the changeover guidelines of plant PL 6, distillation tank levels
and the existing demand requirements, the products P2 and P6 are selected in the rst run for processing in
BILP model. The optimum ow rates across all the arcs obtained from the solution of the FNO model are used
in the BILP model of run-1. Tank levels are initiated/updated for run-1. The generated output of the BILP

1222

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

Table 4
Results of the BILP model for a sample month
Runs

Products selected

Running time of plants (h)

Quantity produced (Cu m)

PL 1

PL 2

PL 3

PL 4

PL 5

PL 6

115
5
96
124
19
50
40
103

160
7
132
123
75
69
55
116

173
116
57
70
11
28
23
148

32
117
87
86
67
70
131
39

59
69
39
34
56
26
62
0

59
67
37
35
57
25
60
4

Total run hours (h)


552
737
626
629
Total quantity produced using BILP model (Cu m)
Monthly demand (Cu m)
Closing inventory (Cu m)
Backorder quantity for month 1 (Cu m)
Actual production quantities observed for a month (Cu m)

345

344

1
2
3
4
5
6
7
8

P2,
P3,
P5,
P4,
P6,
P5,
P3,
P1,

P6
P1
P2
P2
P2
P2
P2
P4

P1

P2

P3

P4

P5

P6

64

3540

4020

3600

2100

2220

1500

3420

64
5000

4936
4176

3540
6670

3130
3456

7620
4300
3320
0
5592

2100
1760
340
0
960

3720
3100
620
0
8064

3420
3150
270
0
1248

Fig. 3. Gantt chart for production plan.


Table 5
Production schedule of plant PL 6 derived from the results of the BILP model
S. no.

Date and time

Product sequence

Flow rate (m3/h)

Yield rate

1
2
3
4
5
6
7
8
9

7/5/06 10:30 p.m.


7/7/06 9:30 a.m.
7/11/06 6:00 p.m.
7/15/06 5:00 a.m.
7/20/06 10:00 p.m.
7/25/06 12:00 p.m.
7/28/06 10:00 a.m.
8/2/06 9:00 p.m.
8/5/06 8:00 a.m.

P2
P3
P5
P4
P6
P5
P3
P1
The time at which the scheduling

60
60
60
60
60
60
60
60
period ends

0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99

model for the run-1 denotes the time for which the selected product is processed by each plant following a
changeover to another product. The results of the BILP model for run-1 are shown in the rst row of Table
4. The subsequent runs are carried out for the planning period of one month. The results of these runs are also
presented in Table 4.

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

1223

Table 6
Results of the evaluative model plant PL 6 schedule
Flow rate m3/h

S. no.

Date and time

Product sequence

1
2
3
4
5
6
7
8
9
10
11
12
13

7/1/06 12:06 a.m.


7/3/06 10:30 p.m.
7/5/06 2:36 p.m.
7/5/06 8:30 p.m.
7/12/06 3:36 a.m.
7/15/06 5:30 a.m.
7/20/06 1:00 p.m.
7/23/06 2:30 a.m.
7/25/06 7:24 a.m.
7/26/06 1:00 p.m.
7/26/06 7:00 p.m.
7/29/06 10:24 p.m.
7/31/06 12:00 a.m.

Product P1
20.25521
Product P2
27.89242
Product P1
29.34056
Product P3
28.21612
Product P2
25.92831
Product P1
20
Product P5
23.28497
Product P6
42.4031
Product P3
36.85575
Product P1
42.36098
Product P3
42.1281
Product P2
37.24661
The time at which the scheduling period ends

Flow tank

Rundown tank

342
341
342
345
341
342
306
340
345
342
345
341

367
366/368
367
361/369/370
366/368
367
316/371/372
360/362/363
361/369/370
367
361/369/370
366/368

Start

Formulation of production scheduling problem for a lube oil section

Setting of objectives for the production scheduling problem

Collection of required data from the lube oil section

Input modeling of the collected data, parameter estimation

Simulation conceptualization and model building of the lube oil section

Initialization of all variables required for the lube oil simulation model

Implementation of appropriate distribution in the simulation model based on the


input modeling
Process each plant stage wise simultaneously and store the quantities produced
in the down stream tanks

When any of the upstream tanks are reaching lower limits, or if any of the
downstream tanks are reaching upper limits, switch over to the next product
based on the changeover guidelines

Compute the quantities produced in each stage and print the required statistics at
the end of simulation

Output analysis of simulation and estimation of confidence intervals

Stop

Fig. 4. Simulation ow chart of the lube oil section.

1224

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

While running the BILP model, one product is selected out of six, and the model is run to satisfy the
demand for that particular product. In this process all the plants are operated to process only the selected
product. Hence, the run time depends upon occurrence of reaching upper or lower limit of any of the tanks
in the selected subsystem (plant). Whenever a plant stops, it denotes that a changeover of product is required
to continue processing. The production quantities computed through BILP model, representative samples of
monthly demand for a particular month, closing inventory, backorder quantities and actual quantities produced for the particular month are presented in Table 4.
5. Deriving production schedule
The results of the BILP model presented in Table 4 can be represented as a Gantt chart and is presented in
Fig. 3. A production schedule for the four plants, namely, PL 3PL 6 are derived from the Gantt chart, and
production schedule for plant PL 6 is presented in Table 5 as a sample. The plant PL 1 handles all six products
simultaneously and plant PL 2 handles only one product. Hence, plant schedules for plants PL 1 and PL 2 are
not considered.
From the results of the BILP model, a production schedule for a time period of one month is prepared. The
deviations of the model results from the actual observed values are due to:
(i) Adjusting feasibility of ow rates across various plants of the lube oil section by the operators based on
limits on tank upper and lower levels;
(ii) Random change of distillate fractions from time to time based on the type of crude and the ratio mix of
dierent types of crude;
(iii) Losses associated with changeovers from one product to another due to process constraints;
(iv) Random process interruptions in the plants;
(v) Operating lube oil plants with reference to demand requirements; and
(vi) Resource interactions.

Fig. 5. Front screen of the lube oil simulation model.

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

1225

If the above variations are absent, the results obtained from the model agree with the design capacity. The
optimization models presented above work on the assumption that resources are available and are managed
optimally. It does not take into consideration the interactions among the resources. Hence, to incorporate the
above deviations in the model, a simulation procedure was introduced based on discrete rate simulation modeling technology. A simulation model was developed based on the discrete rate simulation incorporating all
the features of lube oil section process conditions. This simulation model can be operated on either external
or reactive scheduling mode. In external scheduling mode, the schedule prepared through other techniques like
MILP, Genetic Algorithm, etc, are used to test their feasibility by computing its measure of performance with
the help of the evaluative model. In the reactive scheduling mode, a reactive scheduling algorithm was used for
automatic scheduling of the lube oil plants. This reactive scheduling algorithm works on the concept of sense
and respond mechanism. The high levels, critical levels and low levels of the storage tanks trigger product
change. The lube plant schedules for plant PL 3PL 6 can be obtained with the help of the simulation model.
The production schedule obtained from the simulation model using reactive scheduling algorithm for plant PL
6 is presented in Table 6. The simulation ow chart is presented in Fig. 4. The front screen of the lube oil simulation model developed using EXTEND simulation software package is presented in Fig. 5. The simulation
model results facilitated in the preparation of weekly schedule of lube oil plants.
6. Integration of generative and evaluative models
For the purpose of production planning and scheduling in the petroleum renery, generative models were
used, namely, ow network optimization and binary integer linear programming, which reect deterministic
features of the system. The stochastic features of the system can be captured in an evaluative model like simulation. The results obtained from the generative model are considered as best set of decisions and these decisions are evaluated with the help of the evaluative model. The results of the generative model obtained from
the FNO and BILP models are evaluated with the help of the evaluative model, namely, the simulation model
of the lube oil section.
The results of the generative model are the optimal ow rates across all the arcs of the network as presented
in Table 3 and lube plant schedules of plants PL 3PL 6. The plant schedule of PL 6 is presented in Table 5.
The plant schedules obtained from the generative model are given as input to the evaluative model to nd the
measure of performance of the lube oil section. The measures of performance obtained from the evaluative
model are shown in Tables 7 and 8. The details of integrating generative and evaluative model of the lube
oil section are shown in Fig. 6. This integration of optimization model and simulation model oers the opportunity to capture the benets of both optimization and simulation models [16,17].
7. Validation of the simulation model
The results of the simulation model obtained from external and reactive scheduling modes are compared
with the actual and presented in Table 9. It is found that the total production quantities estimated using
Table 7
Results of the evaluative model as a performance report
S. no.

Performance measures

Value

Notes

1
2
3
4
5
6
7
8
9
10

Total quantities delivered


Total no. of changeovers
Throughput
WIP
Gross prot estimated
Slop produced
Continuous operations index
Finished oils in the nal stage tanks
Total input to system
Simulation period

26,358 Cu m.
28
24,342 Cu m
25,122 Cu m
220.79 Rs in millions
770 Cu m
56%
28,432 Cu m
82,144 Cu m
720 h

Total quantity in Cu m supplied against orders


Lower changeovers reduce the cost and improves the quality
Plant PL 6 determines system throughput in Cu m
WIP held in distillate, ranate and dewaxed tanks
Sales direct raw material costs total changeover cost
In plant PL 5 and plant PL 6 only
% Combined satisfaction of six lube plants
Oils in the nished tanks ready for supply
Total input to plant PL 1
Hours of simulation run

1226

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

Table 8
Results of the evaluative model breakup of the throughput
S. no.

Product

1
2
3
4
5
6
7

Product
Product
Product
Product
Product
Product
Slop

Quantity produced (Cu m)


P1
P2
P3
P4
P5
P6

2966
4199
7291
2230
3240
3989
427

Total

24,342

Generative model : OPTIMIZATION

Evaluative model : SIMULATION


Inputs parameters
External
Schedule

Flow rates of
each arc with
upper and lower
limits

Selection of plant shutdowns


(Planned / Stoc)

Plant schedule

Determination of tank capacity


Product to be
selected for
processing

FNO
model

Overall processing Report

Selection crude Mix

Maximum flow
possible through the
network

Optimal flow
rates across
each arc

Outputs

Selection of external or
reactive schedule

U
L

Changeover cost

Optimal time
of operation
of each plant

Raw mtl cost & Sales price


Demand details
Transportation mode pumping rates

MILP
model

Min processing time of each plant


Product

changeover

time

and

slop

Yield Rate
Flow rate (Fixed / Variable)
Optimal flow rates across each arc

S
I
M

Inputs variables
Starting product

A
T
I
O
N
M
O
D
E
L

i) Total delivered
ii) Total nos of changeovers
iii) Throughput of PL 6
iv) WIP
V) Gross profit
Vi) Gross margin
Vii) Slop produced
Viii) Continuous operating index
iX) Finished lube oils
X) Total input to the system
Quantity of 6 products produced
Plant schedule
Dispatch details
Plant operation report
Satisfaction status report
Changeover cost details
Ending tank levels, Utilization
Inflow report of each plant

Starting tank levels


Simulation run length

Fig. 6. Structure of integrating generative and evaluative model.

Table 9
Product wise comparison between actual, external and reactive quantities produced per month
S. no.

Product

Actual produced (Cu m)

Simulation estimates
External schedule (Cu m)

1
2
3
4
5
6
7

Product
Product
Product
Product
Product
Product
Slop

Total quantity (Cu m)

P1
P2
P3
P4
P5
P6

Reactive schedule (Cu m)

4176
3456
5592
960
8064
1248
480

2966
4199
7291
2230
3240
3989
427

7262
3152
6726
1787
3273
3912
500

23,976

24,342

26,612

J.N. Abraham, K.S. Rao / Applied Mathematical Modelling 33 (2009) 12131227

1227

external scheduling mode is close to the actual production, and statistically the dierences are insignicant.
Hence, we can conclude that the lube oil simulation model is able to imitate the actual operations of the lube
oil section. From the results of the simulation experiments we also conclude that there is a scope for improving
the throughput of the lube oil section using the reactive scheduling algorithm.
8. Conclusions
In the study presented in this paper, the lube oil plants operations are represented as a ow network model.
From this model, optimal ow rates across the network were obtained. A Binary Integer Linear Programming
(BILP) model was developed to nd the optimal time of operation of each plant in the lube oil section to meet
the demand. A production schedule for the lube plants was derived from the results of the BILP model. In
order to incorporate the random variations observed in the lube oil plants, an appropriate simulation model
of the lube oil section was developed [18]. This lube oil simulation model was embedded with a reactive scheduling algorithm which facilitated in deriving the lube plant schedules. This simulation model could be operated
under external schedule conguration to estimate the measure of performance for a given plant schedule. Integration of BILP and simulation models was accomplished to feed the outputs of BILP as inputs to the simulation model. This integration helped to estimate the performance of the lube oil section under various
optimal congurations and also facilitated in planning and scheduling of lube oil plants more eectively.
References
[1] L.F.L. Moro, Process technology in the petroleum renery industry current situation and future trends, Comput. Chem. Eng. 27
(2003) 13031305.
[2] J.M. Pinto, M. Joly, L.F.L. Moro, Planning and scheduling models for renery operations, Comput. Chem. Eng. 24 (2000) 2259
2276.
[3] H. Katayama, On a two-stage hierarchical production planning system for process industries, Int. J. Prod. Econ. 44 (1996) 6372.
[4] Gothe-Lundgren Maud, Jan T. Lundgren, Jan A. Persson, An optimization model for renery production scheduling, Int. J. Prod.
Econ. 78 (2002) 255270.
[5] A. Artiba, F. Riane, An application of a planning and scheduling multi-model approach in the chemical industry, Comput. Ind. 36
(1998) 209229.
[6] J.C. Liza, J.M. Pinto, Optimal scheduling of a lube oil and paran production plant, Comput. Chem. Eng. 29 (2005) 13291344.
[7] Z. Jia, M.G. Ierapetritou, Ecient short-term scheduling of renery operations based on a continuous time formulation, Comput.
Chem. Eng. 28 (2004) 10011019.
[8] G. Xiong, T.R. Nyberg, Push/pull production plan and schedule used in modern renery CIMS, Robotics Comput. Integrat.
Manufact. 16 (2000) 397410.
[9] A. Alle, J.M. Pinto, Mixed-integer programming models for the scheduling and operational optimization of multiproduct continuous
plants, Ind. Eng. Chem. Res. 41 (2002) 26892704.
[10] A. Potter, B. Yang, C. Lalwani, A simulation study of dispatch bay performance in the steel processing industry, Eur. J. Oper. Res.
179 (2006) 567578.
[11] G. Chryssolouris, N. Papakostas, D. Mourtzis, Renery short-term scheduling with tank farm, inventory and distillation
management: an integrated simulation- based approach, Eur. J. Oper. Res. 166 (2005) 812827.
[12] L.F.L. Moro, J.M. Pinto, Mixed-integer programming approach for short-term crude oil scheduling, Ind. Eng. Chem. Res. 43 (2004)
8594.
[13] K. Neumann, C. Schwindt, N. Trautmann, Scheduling of continuous and discontinuous material ows with intermediate storage
restrictions, Eur. J. Oper. Res. 165 (2005) 495509.
[14] R. DeMatta, T. Miller, Production and inter-facility transportation scheduling for a process industry, Eur. J. Oper. Res. 158 (2006)
7288.
[15] C.A. Floudas, X.C. Lin, Continuous-time versus discrete-time approaches for scheduling of chemical processes: a review, Comput.
Chem. Eng. 28 (2004) 21092129.
[16] R. Suri, An overview of evaluative models for exible manufacturing systems, Ann. Oper. Res. 3 (1985) 1321.
[17] K.E. Stecke, Using mathematics to solve problems in industry, INFROMS Trans. Educat. 5 (2005) 18.
[18] J.N. Abraham, Optimization and Simulation in the Management of Process Plant Operations of Petroleum Renery, Unpublished
Doctoral Dissertation, Department of Industrial Engineering, College of Engineering, Guindy, Anna University, Chennai, Tamil
Nadu, India, 2007.