Transportation Analytics — Stories

AMERICAN AIRLINES — Aircraft Maintenance Routing

At American Airlines, thousands of flights are operated daily. Historically, the sequence of flights to be operated by a given aircraft over a schedule month was developed manually. The key requirement that had to be satisfied by a given sequence of flights, known as an aircraft routing, was that it needed to spend 8 hours overnight at one of the designated maintenance locations every 3 or 4 days so that maintenance checks could be performed. As the fleet became larger and larger, this became more and more difficult to develop manually.

To address this issue, an optimization model was developed to solve this problem. Due to the vast number of possible routings, a series of set partitioning problems were developed and solved, yielding "good" routings, where good was defined as:

1. Satisfied the maintenance requirement
2. Operated all the flights with the given aircraft fleet
3. Allowed the desired "through" flights to be operated
4. Provided consistent routings from one day to the next; for example, if Flight XXX connected to Flight YYY on Friday, then it should also do so on other days of the week whenever possible.

Using this model allowed the time to develop a solution to be reduced from 2 weeks to 2 days. This model was later implemented in the Airline Scheduling software package marketed by Sabre under the name, AirFlite.

CONRAIL — Network Analysis Model

Conrail, a freight railroad that operated in the Eastern US, needed the ability to test out operating plan changes analytically before putting them into place. They needed to understand the impact of plan changes on such measures as the number of train starts, number of crew starts, locomotive requirements, train size, delivery schedules, and extra sections. Because of the network impacts of any operating plan change such as changing the time a train operated or changing the block routing to be followed to get from an origin to a destination, it was not possible for a human to work through all of the impacts of operating plan changes.

To provide the needed capabilities, the Conrail Network Analysis Model (CNAM) was developed. CNAM ran a multi-day deterministic simulation to flow the inputted freight across the network using the operating plan that was also provided as an input. Parameter-driven rules provided guidance so that if trains had too many cars scheduled to them, the train could only run up to its allowed length and cars would be left behind to be re-scheduled. The model was written in C++ with links to an Oracle database where all the events were cataloged. Since the results were kept in a relational database, any desired report could be created, including reports that compared the events of one "scenario" to another. Conrail used this tool for many years to evaluate proposed operating plan changes.

CSX — Plan Integrity

CSX, a freight railroad operating in the Eastern US, had several systems in which they made changes to their operating plan daily. Often, because these systems operated independently, changes would be made to one aspect of their operating plan that made it inconsistent with another aspect of their operating plan. For example, an analyst could make changes to the blocks maintained at a given terminal by adding a destination block to the operating plan for that terminal. However, since the system where the assignment of blocks to trains was separate, that new block would not automatically be assigned to a train.

While normally the number of changes being made to an operating plan on a daily basis is not significant, CSX was making major wholesale changes to its plan as part of the Conrail acquisition process. In order to quickly identify inconsistencies in the components of the operating plan, the Plan Integrity system was developed. In this tool, data extracts from the various systems are loaded into a central database, and then a sequence of analyses are performed with reports generated to quickly identify potential problems. Some of these reports and analyses are straight-forward while others employed complex logic.

The Plan Integrity tool was developed in 1998 during the Conrail acquisition and is still in use at CSX.

BAX GLOBAL — Shipment Routing Model

BAX Global operates a freight distribution network using both aircraft and trucks to move large-size packages. They needed a tool to be able to evaluate different flight schedules and truck networks in order to size their fleet, develop accurate budgets, and provide routing guides. The shipment routing problem was formulated as a large-scale mixed integer programming problem where the columns represented potential shipment routes and the rows represented the aircraft or truck being used to provide transportation.

Data were input via Excel spreadsheets into an Access database, and C++ code was executed to generate the mathematical representation of the problem. This problem was then fed into CPLEX which was used as a solver. Results were then fed back into Access for reporting purposes. Users could easily make changes to a scenario by revising either the Excel spreadsheet or the Access table that represented an aspect of a scenario. For example, the user could edit the shipment table, the table representing the flight schedule, the aircraft capacities (both weight and volume), or the costs, and then rerun a scenario. Since the data and the results were stored in Access, it was also easy to compare results of different scenarios.

PORT OF LONG BEACH — 20 Year Plan Support

The Port of Long Beach operates one of the largest seaports in the country if not the world. In 2001, as part of a 20-year planning project, they wanted to project the demand for rail capacity, both on dock and off-dock, so they could plan accordingly. Using historic data and forecasted growth rates, we developed projections of inbound volume by destination and steamship line. Using this, we were able to build a "Train Generating Function" that identified train volumes at various levels of aggregation. This was then used to estimate the number of on-dock and off-dock train starts by day given different demand projections.

PACER STACKTRAIN — Empty Repositioning Optimization

At one time, Pacer Stacktrain had the largest domestic container fleet in North America. In order to more effectively manage this fleet, they needed to be able to determine empty repositioning needs in an optimal manner rather than in a reactive mode. We developed a network optimization model that, using a time-space network, determined the most cost effective set of empty repositioning moves over a multiple-month time horizon. The goal was not only to reduce empty repositioning costs, but also to make containers more available so that additional revenue loads could be handled with the same fleet. A set of inputs were managed via Microsoft Access with the optimization performed using a CPLEX MIP solver. The output was then directed into MS Access so that reports could be generated. The model identified several opportunities to change the way that empties were handled and was able to increase the carrying capacity of the existing fleet.