Production Planning Via Integer Linear Programming
Production Planning Via Integer Linear Programming
Contents
1 Executive Summary
2 Company Background
3 Problem Statement
4.1
Our Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2
4.2.1
Variable Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.2
Parameter Definitions . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.3
Objective Function . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2.4
Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.1
General Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.2
10
5.3
11
6 Limitations
14
15
Executive Summary
J.P. Molasses manages a complex supply chain of raw sugar suppliers, major refineries,
and end customers. We were tasked with providing actionable information to optimize their
supply chain for maximum profit. In order to model the supply chain, we developed a linear
program that took into account all of the constraints (demand, finished good yields, refinery
capacities, etc..) and produced an optimal solution and highlighted the conditions under
which the solution remains optimal.
After running our model, we determined the four most defining characteristics of the
current state of J.B. Molasses supply chain.
1. J.B. Molasses refining operations are capable of providing molasses to company owned
subsidiaries at no cost while simultaneously earning a profit ($69,026.33) as an independent business operation.
2. The Charleston refinery handles a much larger portion of the company business than
the Atlanta refinery, but the Atlanta refinery will perform a useful role as long as
Walsh remains a large customer.
3. The Charleston refinery is producing refined sugar at maximum capacity. Upgrades
to the refinery increasing this capacity an additional 302 K Lbs / month will make
for significant profitability and operational impact on the refining process.
4. Before the Charleston refinery is upgraded, additional supplies for the refining process
must be sourced. Our analysis will remain accurate until an additional 72,000 Lbs is
added to the supply chain. Once more than 72,000 Lbs in additional raw sugar can
be reliably secured, the model will need to be modified and run again.
Perhaps most importantly, although our model guarantees mathematical optimality, our
assumptions should be well understood before its results are applied. Those tasked with
implementation will want to pay careful attention to Sections 6 and 5.2. Detailed results
are contained in Section 5.1.
Company Background
J.P. Molasses was founded as a rum distillery in 1872. Because rum is made from
sugarcane and sugarcane byproducts (most notably molasses), the companys profits were
closely tied to the price of these commodities. In 1953, company leadership made the
decision to build and operate its own refineries in order to insulate J.P. molasses from
significant price hikes occurring in the sugar refining industry. Building these refineries was
a financial blessing and an operational curse. Although they allowed the company to better
control production and avoid middlemen, they also introduced three major questions.
Problem Statement
Cyril Morel provided us with the following directive,
Viewing the refineries as profit centers, maximize profit subject to constraints
given by refinery managers, and revenue information supplied by accounting.
Meanwhile, refinery managers explained that there were three types of constrains: demand
constraints, operational losses and yield constraints, and capacity constraints. Before proceeding with our mathematical model, our team confirmed that the following needed to be
satisfied.
Capacity Constraints
If open, each refinery must operate between 50 percent and 100 percent of its maximum
capacity. Additionally, the Charleston refinery cannot produce any more than 2,250 K Lbs
of purified sugar.
Our model includes an 8th customer. The 8th customer reflects the open market and acts as limitless
source of demand.
Refineries incur both fixed and marginal costs. Charleston costs $10,000 for every month
that it is open, while Atlanta costs $14,200 for every month that it is open. Charleston has
a marginal cost of $31 per 1,000 Lbs raw sugar, while Atlanta has a marginal cost of $30
per 1,000 Lbs of raw sugar.
4.1
Our Approach
In order to model the problem, we developed a linear program (explained more in depth
in the next section) that took into account all of the individual factors (cost, percentage
loss, etc.) in order to maximize the profit for J.P. Molasses. First we define variables for
each decision in the problem. Then we construct a maximizing function that takes into
account all of the costs (transportation, fixed production, etc.) and subtracts it from the
gross revenue. Lastly, we developed constraints that covered all of the restrictions for each
part of the production process. Readers familiar with Operations Research should note that
our model has more constraints than is necessary to achieve the correct optimal solution.
This was done intentionally to increase the detail of the sensitivity analysis.
4.2
4.2.1
Parameter Definitions
max
Objective Function
hX
i
Profit =
(fi,Charleston + pi )Xi,Charleston + (fi,Atlanta + pi )Xi,Atlanta
iSuppliers
hX
i
[10000QCharleston + 14200QAtlanta ]
hX
i
+ 36
YCharleston,k + YAtlanta
kPaying Customers
+ 200WCharleston,sugar + 150WAtlanta,sugar
+ 25WCharleston,bagasse + 25WAtlanta,bagasse
4.2.4
Constraints
i Suppliers
Rj cj Qj
j Refineries
j Refineries
l Products
Charleston cannot exceed its maximum capacity for refined sugar production.
WCharleston,sugar 2250
Refineries supply of molasses cannot be overdrawn.
P
kCustomers Yj,k Wj,molasses j Refineries
Demand for each molasses customer must be satisfied.
YCharleston,k + YAtlanta,k Dk
k Customers
5.1
General Results
Table 2 gives an immediate picture of the expected finished products for both refineries.
Table 1 answers the more pressing question of which customers each refinery will supply.
Table 3 describes how both refineries will be supplied. Note that in only one instance do two
refineries supply a single customer. Note also the fact that Charleston dominates Atlanta
in terms of how much of the raw sugar supply it consumes. This is because Charlestons
sugar (the primary source of revenue for the refineries) is significantly more valuable than
Atlantas. Moreover, shipping costs from Charleston are, with few exceptions, less than
shipping costs from Atlanta.
Refinery
Atlanta
Charleston
Customer
Ashberry
Casey
Market
Quentin
Shorewood
StMarys
Walsh
Wilksboro
Ashberry
Casey
Market
Quentin
Shorewood
StMarys
Walsh
Wilksboro
Total
K Lbs Molasses
850
0
0
0
0
0
948.464
0
0
575
132.179
480
107
80
21.5355
640
3834.1785
Revenue ($ USD)
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
4758.44
0.00
3852.00
2880.00
0.00
0.00
8610.44
Refinery
Atlanta
Charleston
Total
Product
bagasse
molasses
sugar
bagasse
molasses
sugar
K Lbs
1388.39
1798.46
2671.33
2410.71
2035.71
2250.00
12554.60
Revenue ($ USD)
34709.75
0.00
400699.5
60267.75
8610.44
450000
954287.44
Supplier
Coleridge
Harvester
Larkin
Milligan
Omega
Portia
Spectra
Thistle
Refinery
Atlanta
Charleston
Atlanta
Charleston
Atlanta
Charleston
Atlanta
Charleston
Atlanta
Charleston
Atlanta
Charleston
Atlanta
Charleston
Atlanta
Charleston
Total
5.2
In linear programming, an optimal basis is the collection of all variables with value
greater than zero. A variable is said to be out of the basis when it is equal to zero. If the
exact solution of a problem changes, but the basis remains the same, then it is short work
to arrive at the new solution. However, when the basis changes, the linear programming
model must be re-evaluated to reflect this new information. Below we give the conditions
for which our model would need to be re-evaluated. We do not speculate on the potential
causes of these changes in constraints or variables, as only the values of the variables will
determine whether the optimal basis has changed. If any of the following occur...
1. Maximum operating levels of the Charleston refinery decrease by more than 1333.57
K Lbs, or 2921.81 K Lbs for the Atlanta refinery.
2. Minimum operating levels of the Charleston refinery increase by more than 2681.43
K Lbs, or 1468.19 K Lbs for the Atlanta refinery.
3. More than 23.57 K Lbs of Charleston sugar becomes unavailable.
10
5.3
Before proceeding with sensitivity analysis, it is important to note that each course
of action described in this subsection is considered in isolation of all others. If multiple
modifications are made to the supply and production system, then the model would have to
be re-evaluated with this new information in mind. That said, we can move on to analysis.
In the following tables, RHSMax is the maximum amount that a constraints Right
Hand Side2 can increase until the basic solution changes; RHSCurrent is the RHS at optimality; RHSMin is the is most that an RHS can decrease until the basic solution changes.
A Shadow Price is the amount by which the objective function is changed for increasing
the RHS of a constraint by one unit (as long as we are within the bounds set by RHSMax
and RHSMin ).
As can be seen in Table 1 Atlanta currently has two customers. Our team suspected
that Atlanta may only have these customers because Charleston hit an upper bound on
production. As Table 4 shows, while Charleston did not hit its total capacity, it did hit
its capacity on raw sugar. Since raw sugar is such a big source of revenue for the refining
process, we suspected that Atlanta only takes on customers once Charleston can no longer
fulfill demand. Although it is true that Charleston hitting its upper bound caused Atlanta
to pick up some remaining demand, this is not the whole story. Our team re-ran the
model without the upper bound on Charlestons sugar production. The effect was to shift
2
assume all variables are moved to the left, leaving only a constant on the right hand side of the constraint.
11
responsibility for Ashberry from Atlanta to Charleston (taking away one customer), but
Atlanta did retain responsibility for Walsh.
Constraint
minimum Charleston
Atlanta
maximum Charleston
Atlanta
max sugar Charleston
Shadow Price
0
0
0
0
2.20494
Slack
2681.43
1468.19
1333.57
2921.81
0
RHSMax
2681.43
1468.19
2551.02
RHSCurrent
0
0
0
0
2250
RHSMin
-1333.57
-2921.81
2226.43
One factor we are very interested in is how the yield proportions for different products
affects the optimal solution. In order to try to get a sense for this, our team included an
equality constraint that reflected the relationship between the amount of a product after
refining (Wj,l ) and the amount of raw sugar coming into a refinery (Rj ). Table 5 does
reflect presence of these yields, but the changes to the RHS of an equality constraint are
not meaningful when both sides of the constraint are constrained by variables instead of
constants.4 In fact, the only way to gain information on how different yields impact the
objective function is to solve the ILP5 model with different yield proportions. Given that
sugar yields are so closely tied to the profitability of the refining process, it may be wise
to run the model with different yield proportions. However, to focus too heavily on the
amount of sugar J.P. Molasses produces would be misguided. Our team may have been
tasked with optimizing a refinery production plan, but we have done so without regard to
J.P. Molasses overall business model.6
One of the more useful pieces of information from sensitivity analysis came from supply
constraints. Table 7 shows that each supply constraint has a slack of zero- that is- that J.P.
Molasses has exhausted all supply. By turning to shadow prices, we are able to identify
the suppliers most valuable to J.P. Molasses given these market conditions. The relative
3
Atlanta continues to supply Walsh (in the amount of 673.426 K Lbs). Note that Walsh is the single
largest customer and also the most difficult to supply.
4
We experienced similar problems when attempting the analyze the constraints the reflected the retention
proportion of raw sugar shipped from supplies to refineries (see Table 6)
5
integer linear programming
6
And undoubtedly, the refining process is just one component of J.P. Molasses revenue generating activities.
12
Constraint
Charleston
Atlanta
Sugar
Molasses
Bagasse
Sugar
Molasses
Bagasse
Shadow Price
197.795
36
25
150
47.3
25
Slack
0
0
0
0
0
0
RHSmax
23.5698
21.5355
RHScurrent
0
0
0
0
0
0
RHSmin
-301.019
-132.179
-2410.71
-2671.33
-132.179
-1388.39
Constraint
To Charleston
To Atlanta
Shadow Price
55.4031
58.8461
Slack
0
0
RHSMax
70.1483
70.1483
RHSCurrent
0
0
RHSMin
-430.55
-430.55
Supply Limits
Thistle
Spectra
Portia
Omega
Milligan
Larkin
Harvester
Coleridge
Shadow Price
25.5073
24.7073
20.2573
26.6073
22.9573
24.4073
29.1573
22.7573
Slack
0
0
0
0
0
0
0
0
13
RHSMax
2072.09
1072.09
2212.09
1442.09
1922.09
1655.09
1332.09
1772.09
RHSCurrent
2000
1000
2140
1370
1850
1583
1260
1700
RHSMin
1557.5
557.503
1697.5
927.503
1407.5
1140.5
817.503
1257.5
Figure 1: Increase in profit per 1,000 Lbs additional raw sugar (up to 72,090 Lbs)
Portia!
Coleridge!
Milligan!
Larkin!
Spectra!
Thistle!
Omega!
Harvester!
0!
5!
10!
15!
20!
25!
30!
35!
Limitations
Significant Limitations
1. In general, our model is static. We used fixed prices for commodities and transportation costs.
2. We assumed that all products sold on the open market (including molasses) incur no
shipping costs.
3. Our model assumes that all steps of the refining process (from sourcing to refining to
delivery) occur with 100% reliability.
More subtle limitations.
1. Our model does not consider the possibility of alternative transportation routes.
2. Our model assumes one month inventory periods, while inventory management would
advise working in smaller timesteps.
3. Coordinating cross-state purchasing in any particular way may have tax ramifications.
Our model assumed that these tax ramifications are negligible.
4. Our model assumes that the amount of a commodity that J.P. Molasses sells on the
open market does not influence the price of that commodity.
14
we would advocate a reduction in molasses production and perhaps even expansion into
sugar refining as a primary business. Because J.B. Molasses refineries sell each product
at market prices (at least, portions of those products), it can easily be seen that sugar
generates significantly higher profit margins than molasses. We abstain from making such a
recommendation because it is important to not lose sight of the fact that the refineries are
a support business. J.B. Molasses is a long standing company with operations that seem
to exhibit a high degree of fit.7 . The value of this fit may or may not have a strategic
value that exceeds the lure of sugar refining as a primary business; this is left the corporate
leadership to determine.
With this critical fact in mind, we recommend the following course of action:
1. Begin searching for additional sources of raw sugar (whether from existing suppliers
or new suppliers, these will eventually need to be integrated into a new Integer Linear
Programming model).
2. While searching, implement the Production Plan given in Section 5.1, and monitor
the system for the conditions in Section 5.2. Expect to generate ($69026.33) in profit
if all prices and costs remain the same.
3. Once the search is complete, run various forms of Integer Linear Programming models
that consider both the new supply and candidate levels of upgrades to Charlestons
refined sugar capacity.
4. Using the results of these models, project cash flows for expected life of the upgrade,
and select the scenario with the highest Net Present Value.
15