Optimal Tank Farm Operation

Sebastian Terrazas-Moreno
Ignacio E. Grossmann
John M. Wassick

EWO Meeting
Carnegie Mellon University
March 2011

In collaboration with The Dow Chemical Company

A tank farm is a set of storage tanks that hold finished
product until it is shipped

Loading of product takes place

Each tank
only from storage tanks
can only
hold one
product

Dedicated
Tanks!

Without available
storage space
finishing lines
cannot process
Material flow diagram used for the tank farm
product

• Unsatisfied product demand results from having no available storage

space at the time scheduled for production
• This is referred to as blocking finishing lines

Page 1
 Dedicated storage tanks are assigned at the beginning of
the operating horizon

Illustrative example

Finishing Existing Tank

C
Connections
ti Farm
Order No. Line Product. Quantity (klb) Li
Lines
Tank 1
1. I A 500 Line I

Shippin
200
2 I B 100 A | B

3 II B 180

ng to custome
Tank 2

P
Process
Line II
4. II A 100 100
B | A
5 III C 450
6 III A 100 Tank 3
Line III

er
C | A
250

Product Production Rate Shipping rate

any line (klb / hr) any tank (klb / hr)
A 50 35
How should we choose which product to store
B 15 7
in each tank?
C 30 15 i.e., determine product – tank assignment

Page 2
 Management of tank farm operation has an impact on the
profitability of the manufacturing process

Profitability driver Impact of tank farm operation

• Unsatisfied product demand can be minimized by the optimal

More product to tank assignment
productivity

• Optimal utilization of available tanks minimizes the need for

additional investment in storage
Less capital
p
investment

• Optimal tank farm management allows the introduction of new

products and the addition of extra volume for existing ones
Better asset
utilization

Page 3
 In practice, the tank farm operation is highly constrained

• Some tanks cannot be loaded and unloaded at the same time

• Th
There might
i h be
b a minimum
i i volume
l ffor unloading
l di a tank k
• There might be a maximum level of loading for safety reasons
• Some tanks are not suitable for storing certain products
• Shipping is not done continuously but in discrete amounts using railcars and
tank trucks
• Shipping can only take place during certain hours

 If all of these constraints are incorporated into an optimization model the

problem becomes huge and hard to solve.
 A discrete event simulation can capture all of these constraints without
becoming computationally intractable
 For this reason we propose a hybrid optimization – simulation approach

Page 4
 The objective is to test the potential of an optimization and
simulation approach for management of tank farms

Current decision making process for Proposed optimization + simulation

product to tank assignment approach

Engineers & managers Run optimization model to obtain

Devellop optimiza
use know-how and experience to tank – product assignment
decide how to assign tanks

A B C A B C

ation model
Tank 1 Tank 2 Tank 3 Tank 4 Tank 1 Tank 2 Tank 3 Tank 4

Use discrete event simulation to

Actual plant operation shows predict performance of the
the performance of the tank farm tank farm

Page 5
 The optimization model is based on a simplified model of
the tank farm operation

Line 1 Tank 1 Given

• A set of products
• A set of production lines
Tank 2
t0 t1 t2 t3 t4 … tn • A set of tanks
• Approduction schedule
Feasible Tank 3 • A set of possible connections
Set of time points ti connections
between lines and storage
Line 2 tanks
Tank 4
• A set of shipping resources
(e.g., railcars)
t0 … tn Tank 5 Determine:
• Assignment of products to
At each time p
point: tanks.
– Read from schedule the product being processed With the objective of:
– Determine how much product can be transferred to tanks • Minimizing unsatisfied quantity
– Calculate mass balance at storage tanks

Optimization model is equation oriented;

it does not include detailed dynamics of tank farm operation Page 6
 The problem is formulated as an MILP

z j
j ,k 1 Assignment constraint 1

z j ,k  cmp j ,k Assignment constraint 2

p
t
j ,k ,l ,t  z j ,k rate j ,l T Knapsack-type inequality;
constraint on transfer rate
s j ,k ,t  s j ,k ,t 1  c j ,k ,t 1 ( t   t 1 )   p j ,k ,l ,t 1 ( t   t 1 ) Inventory balance at time points
l

s j , k ,t  v k Storage constraint

c
k
j ,k ,t  shh j Shipping constraint

ua j  tgt j     p j ,k ,l ,t ( t 1   t )  0 Unsatisfied demand

k l t

max obj   (tgt j  ua j ) Objective function:

j Max satisfied demand =
Min unsatisfied demand

Page 7
The optimal product – tank assignment is used as input in
a detailed discrete-event simulation

The simulation model requires a fixed product tank assignment

Flow logic for discrete event simulation model

Page 8
 We tested the approach by using data from a real process

Get data from Process:

• Schedule for 60 finished products
• Production capacity of 8 finishing lines
• 80 tanks of different sizes and load/unload
characteristics
• Finished product shipping modes and rates1
Compare performance
of optimal assignment vs.
Run optimization model with given data to obtain tank assignment provided
optimal2 product
product-tank
tank assignment
assignment. This step leads to by business (obtained
a problem with millions of constraints and continuous without optimization)
variables and thousands of discrete variables. using existing Arena@
Solution times are in the order of hours. simulation model

Run discrete-event simulation model with optimal

product-tank assignment
p g to p
predict performance
p
(1) Shipping rates are approximate.
(2) It is rigorously optimal only for the simplified representation of the tank farm operation used in the
optimization model Page 9
 The simulation shows that the tank farm operation can be
improved by exploiting the optimization tool

Total shipments, unsatisfied quantity and orders received in 1 year. Simulated with
business provided assignment and optimal assignment
Mass units

Business product-tank assignment Optimal product-tank assignment

1 1
1.01
35% 0.35 0.93
28% 0.25

Mass 0 65
0.65 Mass 0 65
0.65
units units

0 0
Shipments Unsatisfied Orders received1 Shipments Unsatisfied Orders received1

(1) Difference in orders received between both assignments is explained as follows: An order is received when either of two conditions
are met: a) production of previous order is finished or, b) previous order is identified as unsatisfied quantity. In business assignment
more orders are considered unsatisfied allowing more total orders to be received. Page 10
 The optimization approach needs to be extended to
account for changing operating conditions

• Optimal assignment
Business product-tank assignment Optimal product-tank assignment
was determined for
1.01
0 93
0.93 constant shipping rates

0.35 0.25
0.65 0.65 • When the same
Constant shipping

Mass assignment is used to

units
35% 28% simulate the system
with different
(stochastic) shipping
rates

rates, performance is
Shipments Unsatisfied Orders Shipments Unsatisfied Orders
no longer better than
business assignment
0.08 0.87 0.11 0.87
0.77 • Optimization model
Stochastic shipping

0.74
needs to contemplate
Mass 10% 12% variability in operating
units
parameters
t (
(e.g.,
stochastic
rates1

programming, robust
Shipments Unsatisfied Orders
optimization)
Shipments Unsatisfied Orders

(1) Stochastic shipping rates are lower on average than the constant rates we used Page 11
 Summary of results

• An optimization model was able to produce a product to tank assignment

that resulted in better operation
p ((less unsatisfied product)
p ) than the
assignment proposed by the business

• The improvement was a result of more efficient utilization of tanks assigned

to high volume lines

• Real process is subject to variable operating conditions. In order to

generate a robust computational tool, variability of operating parameters
must be considered in the optimization model

Page 12
 Next steps

• Include possibility of delaying production orders:

Scheduling of finishing lines

• Include stochastic elements in optimization model:

Variable shipping rates

• Variable number and size of additional storage tanks:

Design as a degree of freedom

Page 13
Thank You

Page 14