This paper describes a simulation model developed as a decision support system for capacity planning of distribution terminals for a Brazilian oil company. Due to the complexity of loading demands, caused by different combinations of products and quantities, simulation was used.
The fuel distribution sector in Brazil has undergone several transformations in recent years. In 1996, this sector handled a volume of around US$ 25 billion, second only to the commerce, automobiles and food sectors. The sector has undergone deregulation in recent years, with the internationalization of the market and the entry of competitors.
To increase logistical efficiency, a distribution company is investing in redefining its logistics network, streamlining its transport system, implementing customer support services and increasing the productivity of its Distribution Bases. It is precisely in this last point where we will present the application of the simulation technique.
DISTRIBUTION SYSTEM
Fuel distribution starts at each of the 13 refineries in the country. The products are transferred and stored at the Distribution Bases, where tank trucks are supplied and mixed with the company's own products. From the Distribution Base, the products are sent to the company's final customers, such as service stations, large consumers and wholesalers. Figure 1 illustrates the company's distribution system.
 |
The flow between refineries and Distribution Bases is predominantly done through pipelines, while distribution from the Base to the end customer is only done by road transport, with most of the company's truck fleet.
This work describes a tool to aid in the sizing of Distribution Bases, minimizing the waiting time for trucks at the Base. The shorter the waiting time at Distribution Bases, the greater the number of trips that trucks can make to the end customer, thus using the company's resources more efficiently.
COMPLEXITY IN SIZING DISTRIBUTION BASES
A Distribution Base, in a simplified way, is composed of fuel storage tanks and bays for loading tank trucks. Most of the trucks are compartmentalized, thus enabling the loading and transport of different types of fuel and quantities. In each service bay, there are loading nozzles for each type of fuel.
Thus, sizing a Distribution Base means determining the number of service bays and the mix of fuel nozzles in each of these bays, that is, what type of fuel the nozzle should carry. Configuration changes are extremely expensive, making it impossible for configuration tests to be done with the real system.
Such dimensioning is not a trivial task, due to the complexity inherent to the demand for shipments. Trucks arrive for loading that is not constant over time, with demand peaks occurring. Graph 1 shows the percentage of truck arrivals by time slot on a given day.
 |
In addition, because they have compartmentalized tanks, each vehicle demands a different quantity and mix of products. Graph 2 shows the loading diversity in a given month. In the case below, 32% of the trucks that entered this base only loaded Diesel (D), 20% loaded Hydrated Alcohol (AH) and Gasoline (G), and so on.
 |
In this way, we have trucks arriving at the Distribution Base at different time intervals, requesting different fuel mixes in different quantities. This complexity in demand makes it difficult to determine the number of bays the base should have and what types of fuel each bay should have. The company then used analytical formulas for sizing, based on the total volume of demand, which made it necessary to use safety factors and oversizing Bases.
THE SIMULATION TECHNIQUE
The technique used in this study was simulation. This approach is primarily about creating a model that represents reality. The model, by properly representing the operation of the real system, after a validation stage, can then be used to test alternatives for operation different from the current ones and compare them to each other. According to Banks, Carson & Nelson (1996) [1], simulation is indicated when the system to be studied is complex, and the relationships between the various variables are difficult to determine or measure.
For the formulation of the problem, the configuration of the bays and the demand profile were defined as control parameters. The output variables were the waiting times for the trucks, the queue time and the use of service nozzles. According to Saliby [2], the model to be described below is a probabilistic, dynamic model with discrete events.
The model was built using the Arena® simulation software. The steps taken to build the model followed the methodology described in Law & Kelton [3]: (a) Formulation of the Problem, (b) Obtaining data and defining the model, (c) Construction of the model, (d) Validation, (e) Definition of experiments and (f) Analysis of Results. Modeling included the use of probabilistic distributions for the arrival rate of trucks and detailing the compartments of each truck, that is, different mix and quantity of products to be loaded. Each bay could have different types of loading nozzles, where each nozzle had a specific flow rate, depending on the type of fuel. In addition to the loading stage, the model included a check-in and check-out process, where security and verification procedures are carried out by the company. The model also took into account truck movement times within the Distribution Base.
The model validation was performed by comparing the information generated with the simulation model with the company's historical data.
RESULTS
We used 3 factors (input variables) for variation in the system: pump flow, number of slabs and number of nozzles. As response variables, we computed Total Load Time and Queue Time. In Graph 3, we have an example of the Total Loading Time for different flow rates.
 |
In a first group of experiments, we observed how the variation in factors influenced loading and waiting times, with the main conclusions being:
1. Small variations in flow rates led to large variations in loading and waiting times.
2. The relationship between the number of slabs and the total service time follows a curve similar to a parabola
3. The concentration of nozzles per slab tends to reduce loading time.
4. A decrease in loading time does not necessarily lead to a drop in total service time.
5. We check the relationship between the number of nozzles and the total system time. For a low number of nozzles, the total service time tends to grow exponentially.
6. linear relationship between the number of nozzles and the use of nozzles
An important result of using the simulation model was to observe that the flow factor is the one where small improvements generate great benefits to the system. In addition, the quantification of the relationships between the number of nozzles and the total time and nozzle utilization responses. In this way, an analysis between service level and resource utilization can be made.
In a second group of experiments, the arrival rate of trucks was equalized, avoiding peak moments. This would represent a proactive attitude by the company in working with shipments with pre-programmed time windows, avoiding the concentration of truck arrivals at certain times of the day. We reached the following conclusions:
1. The homogenization of demand substantially reduces the Total Service Time and Waiting Time for trucks at the Distribution Base, in addition to reducing its variability.
2. Changes in truck arrival time do not change loading time, as long as the number of trucks, volume and fuel mix of trucks arriving at the Base remain constant.
CONCLUSION
The simulation was an adequate tool to deal with the complexity of the problem, reaching the level of detail of the compartmentalization of each truck, the variable arrival rates and the different quantities and mix of products.
It made it possible, without investments or changes to the real system, to assess the future impact of varying factors, reaching the level of detail on the use of each resource and the level of service provided to the trucks in the fleet.
It made it possible to quantify relationships between the number of nozzles and their response in service time and usage of nozzles.
It made it possible to evaluate the impact of scenarios that cannot be tested in the real system, such as how the variation in the demand profile of trucks affects the truck service system.
In this way, the use of simulation is a promising approach in the treatment of this problem, making it possible to model a wide variety of configurations and alternative scenarios, in addition to directing investment policies that improve service in distribution bases.
BIBLIOGRAPHY
BANKS, J.; CARSON, J.; NELSON, B. Discrete-Event System Simulation, New Jersey: Prentice-Hall, Inc., 1996
SALIBY, E. Rethinking Simulation – Descriptive Sampling, São Paulo: Editora Atlas, Rio de Janeiro: Editora da UFRJ, 1989
LAW, AM; KELTON, WD Simulation Modeling and Analysis, New York: McGraw Hill, 1991
