Smart loop: simulation improves passenger travel between cars, planes.

PHOENIX SKY HARBOR INTERNATIONAL Airport is one of the busiest airports in the United States and the third largest in terms of rental car transactions, with over 1.5 million per year.

[ILLUSTRATION OMITTED]

In January 2006, a consolidated rental car center opened at the airport to simplify the car rental process. Upon leaving the airport, passengers board a bus that takes them to a facility containing all major rental car companies. This reduces confusion and improves efficiency.

The rental car center contains a 125,000-square-foot customer service building situated above a three-level parking garage. The garage holds up to 5,600 vehicles for 11 rental car companies.

Anticipating greater air travel across the board, the airport sought to increase the capacity of the rental car center. Transportation consulting services firm TranSystems developed several layout alternatives for the project and together with the airport and rental car companies, reduced the number of potential layouts to three, which are shown in Figure 1.

As presented in all layouts, buses enter on the upper left side and follow a counterclockwise flow. In layout F, arriving buses may park in one of the center spots to drop off and pick up passengers. Upon leaving the center island, the bus may turn right to exit the system or left to loop back through the main circle and pick up additional waiting customers. Layout F provided the rental car center with three additional bays for buses to drop off and pick up passengers.

The flow of layout G is very similar to F, but with two center islands instead of one. Layout G also adds four additional bays instead of three. Both layouts G and F add crosswalks. This inserts additional complexity to the system since buses must wait on pedestrians and pedestrians must wait on buses. This also added a quantitative concern since the safety of pedestrians must be considered.

[ILLUSTRATION OMITTED]

The final alternative is B, which adds a peninsula shape and three additional bays. It may look like more bays are added, but some of the center bays are lost at the base of the peninsula. This layout has no pedestrian crosswalk. However, at the peak of the peninsula, there was a high potential for bus congestion, which could impact maximum throughput.

Because of the dynamic nature of the system, it was difficult for the team to determine which layout would increase throughput the most.

The primary objective of this project was to determine which alternative provided the highest passenger throughput capability. An increased passenger throughput rate results in a higher number of passengers that can be processed without incurring excessive wait times. While the rental car center wanted to quantify the impacts of layout on throughput, the quantitative results of the simulation study were only part of their decision-making process. Another major key factor was shareholder support for the recommended solution.

Solution: Unify layouts

An initial model framed the current system using discrete-event simulation software. The model was developed with an Excel software user interface that allowed the user to input key model parameters, which included:

* Bus arrival volumes at the center

* Number of passengers per bus

* Returning passenger arrivals at the center

* Time required to load and unload buses

* Passenger walk times

* How buses are assigned to bays

The input table allows the user to define the number of buses being used by time of day as well as how these buses were distributed to the various terminals over an entire 24-hour period.

To populate the required inputs, the rental car center provided historical data of passenger volumes by terminal and time of day. Time studies were conducted to determine passenger load and unload time distributions.

[FIGURE 1 OMITTED]

The model used input parameters from Excel and launched the dynamic model, which could be run with or without animation. Once the model has run, outputs are loaded into the Excel user interface and the user is provided with summary tables and graphs. Figure 2 demonstrates how the buses are processed per hour by the rental car center and the total number of buses in the system over the course of a 24-hour period.

Being able to examine system behavior over time is an advantage of using dynamic modeling for a design problem like the rental car center bus loop. Other outputs included bus time in system, passenger wait times, passengers processed per hour, and crosswalk-related statistics for systems with crosswalks.

Validation: Prove logic

After the baseline model was developed, a validation effort was conducted to ensure the model adequately represented reality. Bus time in the system was defined as time from entering the rental car center bus loop to leaving the loop, and time stamps for an entire day's worth of buses were available from the real system. This time stamp would cover bus travel time within the rental car center, pulling into and out of bays, passenger loading and unloading times, and any bus waiting times. This was essentially all of the major system components modeled for the study. Passenger volumes and bus schedules from that day were entered as inputs into the model, and then the bus time in the system from the real day were compared to the output results of the model.

A two-sample pooled t-test was used to compare the time in the system of the actual system to the simulation model. Initial hypothesis testing showed that the model did not predict the performance of the system to the desired level of accuracy. This resulted in the team re-evaluating some of the simplifying assumptions and verifying input parameters.

[FIGURE 2 OMITTED]

More detailed distributions for bus time at the terminal were added to reflect bus arrival patterns accurately. Sometimes a bus would stop at two bays to pick up passengers. Because the bus was already partially full at the second stop, this slowed down the loading process at that stop. This detail was added to the model to improve accuracy. After these model changes, the model predicted actual system performance with the required accuracy of 90 percent.

The model analysis produced unexpected findings. After validating the baseline model, the additional three layout alternatives were added to the model. The user could evaluate layout performance by selecting which layout to use with the Excel user interface. Based on the baseline validation, the team was confident that the model accurately modeled bus and passenger behavior. The only factor now varied was the layout.

Given that maximum passenger throughput was the primary metric of interest in determining the best layout, the model inputs were artificially set to have an extremely high number of passengers in the system. Having plenty of passengers ensures buses never leave or arrive into the model partially full.

The simulation analysis showed none of the alternative layouts as better than the current layout. It is counterintuitive that additional bays do not improve system performance. The team's first response was to re-check the validation effort and model inputs. After finding them to be correct, the team began to evaluate if all three of the layout alternatives and the current layout really could produce the same maximum throughput.

[FIGURE 3 OMITTED]

They brainstormed the factors in the system that would most likely affect passenger throughput. The final list included the layout (number of bus bays), how the system was managed, and the number of buses in the system. The system can follow several different rules for managing the flow of buses. This includes whether bays will be dedicated to all pick up or all delivery or if they can act as a dual pick-up and drop-off. In Figure 3, the red vertical lines indicate when all bays perform in this dual mode.

This graph shows an increase in maximum throughput during these dual mode times and indicates this approach does have a positive affect on throughput. Additional experimentation showed this approach only added throughput during peak times. During slow times, this approach had no significant affect on throughput or wait times.

While the change in operating procedure does influence maximum passenger throughput, it does not explain why all the layouts performed the same. The team began to examine whether the number of active buses in the system was the limiting factor that caused all layouts to perform the same.

Figure 4 illustrates maximum passenger throughput with varying numbers of buses in the system for the current layout. Increasing the number of buses increases the maximum throughput for the current layout. The increased throughput seen during the middle of the day reflects when the bus-to-bay assignment is managed differently. This finding about bus-to-bay assignment is being used by the rental car center to manage the system during peak times better.

In Figure 4, the system performance is the same for 92 buses and 102 buses. This is the point where the system bottleneck shifts from the number of buses in the system to the layout, or number of bus bays, in the system.

This analysis found that with an unlimited number of buses in the system, some layouts perform better than others. While increasing the number of bays can increase throughput, this increase is only achieved if additional buses are added. The number of buses in the system, not the number of bays at the rental car center, is placing the largest constraint on system throughput.

[FIGURE 4 OMITTED]